[c-nsp] Cisco MultiLink PPP

[Lawrence Wong]

Hi all,

I am thinking of combining 2 x T1 (1.5Mbps) into a
larger pipe of 3Mbps for greater capacity via MLPPP.

I would like to check if MLPPP is supported on all
Cisco platforms with Serial interface or is it only
supported on selected platforms.

Under MLPPP, will traffic be distributed amongst both
links (i.e. I can download at max of 4Mbps) or am I
limited to the max of each link (i.e. I can download
at max of 2Mbps).

How processor intensive is MLPPP?

Are there any caveats to look out for?

Does anyone have experience with MLPPP to share?

TIA!

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Oliver Boehmer (oboehmer)]

Post by Lawrence Wong
I am thinking of combining 2 x T1 (1.5Mbps) into a
larger pipe of 3Mbps for greater capacity via MLPPP.
I would like to check if MLPPP is supported on all
Cisco platforms with Serial interface or is it only
supported on selected platforms.

It is supported on all low- to mid-end routers up to and including 7500,
check the release notes and/or feature navigator for higher-end as well
as hardware-based platforms

Post by Lawrence Wong
Under MLPPP, will traffic be distributed amongst both
links (i.e. I can download at max of 4Mbps) or am I
limited to the max of each link (i.e. I can download
at max of 2Mbps).

A single session can use up to 4Mbps. MLPPP will distribute the packets
over both links making sure to preserve the packet sequence.

Post by Lawrence Wong
How processor intensive is MLPPP?

In halfway recent code it is CEF switched and 2x2Mbps should be a
non-issue.. You can optimize by disabling fragmentation, but it should
not be needed to do this.

Post by Lawrence Wong
Are there any caveats to look out for?

you want to make sure the delay over both links is somewhat equal.
Otherwise MLPPP could run out of buffers trying to reassemble fragments
or preserving the packet order.

oli

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Mohacsi Janos]

Post by Oliver Boehmer (oboehmer)

Post by Lawrence Wong
I am thinking of combining 2 x T1 (1.5Mbps) into a
larger pipe of 3Mbps for greater capacity via MLPPP.
I would like to check if MLPPP is supported on all
Cisco platforms with Serial interface or is it only
supported on selected platforms.

It is supported on all low- to mid-end routers up to and including 7500,
check the release notes and/or feature navigator for higher-end as well
as hardware-based platforms

Except Cisco 8xx. We have serious problems with Cisco 831 with multilink
PPP....

Regards,

Janos Mohacsi
Network Engineer, Research Associate
NIIF/HUNGARNET, HUNGARY
Key 00F9AF98: 8645 1312 D249 471B DBAE 21A2 9F52 0D1F 00F9 AF98

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Gert Doering]

Hi,

Post by Mohacsi Janos
Except Cisco 8xx. We have serious problems with Cisco 831 with multilink
PPP....

OK, I bite. What sort of multilink links are you doing on the 831,
considering that the box only has 2x Ethernet?

I assume that you do things like MPPP over 2x PPPoE or such, and I have
no idea how the resulting config might look like :-) (I know how to do
it for boxes with one ethernet per PPPoE session, but the 831 doesn't
have enough).

gert

[Paul Stewart]

Whew! Someone else asked the question... I had a couple of questions
posted really that were "blonde moments".. was afraid I might be having
another moment...hehehe

We've ended up using 1700, 2600's etc. to get enough ports so really
curious..

Paul


Gert Doering wrote:
| Hi,
|
| On Thu, Mar 24, 2005 at 01:48:48PM +0100, Mohacsi Janos wrote:
|
|>Except Cisco 8xx. We have serious problems with Cisco 831 with multilink
|>PPP....
|
|
| OK, I bite. What sort of multilink links are you doing on the 831,
| considering that the box only has 2x Ethernet?
|
| I assume that you do things like MPPP over 2x PPPoE or such, and I have
| no idea how the resulting config might look like :-) (I know how to do
| it for boxes with one ethernet per PPPoE session, but the 831 doesn't
| have enough).
|
| gert
|

[Lobo]

MLPPP is also supported on the ESR 10008 series.

Jose

Post by Oliver Boehmer (oboehmer)

Post by Lawrence Wong
I am thinking of combining 2 x T1 (1.5Mbps) into a
larger pipe of 3Mbps for greater capacity via MLPPP.
I would like to check if MLPPP is supported on all
Cisco platforms with Serial interface or is it only
supported on selected platforms.

It is supported on all low- to mid-end routers up to and including 7500,
check the release notes and/or feature navigator for higher-end as well
as hardware-based platforms

Post by Lawrence Wong
Under MLPPP, will traffic be distributed amongst both
links (i.e. I can download at max of 4Mbps) or am I
limited to the max of each link (i.e. I can download
at max of 2Mbps).

A single session can use up to 4Mbps. MLPPP will distribute the packets
over both links making sure to preserve the packet sequence.

Post by Lawrence Wong
How processor intensive is MLPPP?

In halfway recent code it is CEF switched and 2x2Mbps should be a
non-issue.. You can optimize by disabling fragmentation, but it should
not be needed to do this.

Post by Lawrence Wong
Are there any caveats to look out for?

you want to make sure the delay over both links is somewhat equal.
Otherwise MLPPP could run out of buffers trying to reassemble fragments
or preserving the packet order.
oli
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Rodney Dunn]

As well as the 76xx with OSM's and Flexwan modules.

Post by Lobo
MLPPP is also supported on the ESR 10008 series.
Jose

Post by Oliver Boehmer (oboehmer)

Post by Lawrence Wong
I am thinking of combining 2 x T1 (1.5Mbps) into a
larger pipe of 3Mbps for greater capacity via MLPPP.
I would like to check if MLPPP is supported on all
Cisco platforms with Serial interface or is it only
supported on selected platforms.

It is supported on all low- to mid-end routers up to and including 7500,
check the release notes and/or feature navigator for higher-end as well
as hardware-based platforms

Post by Lawrence Wong
Under MLPPP, will traffic be distributed amongst both
links (i.e. I can download at max of 4Mbps) or am I
limited to the max of each link (i.e. I can download
at max of 2Mbps).

A single session can use up to 4Mbps. MLPPP will distribute the packets
over both links making sure to preserve the packet sequence.

Post by Lawrence Wong
How processor intensive is MLPPP?

In halfway recent code it is CEF switched and 2x2Mbps should be a
non-issue.. You can optimize by disabling fragmentation, but it should
not be needed to do this.

Post by Lawrence Wong
Are there any caveats to look out for?

you want to make sure the delay over both links is somewhat equal.
Otherwise MLPPP could run out of buffers trying to reassemble fragments
or preserving the packet order.
oli
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Yury Yaroshevsky]

Post by Lobo
MLPPP is also supported on the ESR 10008 series.

How many E1 may be aggregated into
MLPPP bundle on 72xx or 75xx.

This page:
http://www.cisco.com/en/US/products/hw/modules/ps2033/products_white_paper09186a
0080091d4b.shtml

tell us about 2-8 E1 channels.

This page:
http://mail.donbass.net/Redirect/www.cisco.com/univercd/cc/td/doc/product/softwa
re/ios120/120newft/120t/120t3/multippp.htm

tell us about aggregation more than 8 channels.

Please, share practical experience on aggregation more than 8 channels?

[McCallum, Robert]

Robert McCallum
CCIE #8757 R&S
01415663448
07818002241

-----Original Message-----
Sent: 24 March 2005 11:34
Subject: RE: [c-nsp] Cisco MultiLink PPP

Post by Lawrence Wong
I am thinking of combining 2 x T1 (1.5Mbps) into a
larger pipe of 3Mbps for greater capacity via MLPPP.
I would like to check if MLPPP is supported on all
Cisco platforms with Serial interface or is it only
supported on selected platforms.

It is supported on all low- to mid-end routers up to and
including 7500, check the release notes and/or feature
navigator for higher-end as well as hardware-based platforms

Post by Lawrence Wong
Under MLPPP, will traffic be distributed amongst both
links (i.e. I can download at max of 4Mbps) or am I
limited to the max of each link (i.e. I can download
at max of 2Mbps).

You missed the start there Oli - its 2 * T1s which equals 3 meg max.

A single session can use up to 4Mbps. MLPPP will distribute
the packets over both links making sure to preserve the
packet sequence.

Post by Lawrence Wong
How processor intensive is MLPPP?

In halfway recent code it is CEF switched and 2x2Mbps should
be a non-issue.. You can optimize by disabling fragmentation,
but it should not be needed to do this.

Post by Lawrence Wong
Are there any caveats to look out for?

you want to make sure the delay over both links is somewhat
equal. Otherwise MLPPP could run out of buffers trying to
reassemble fragments or preserving the packet order.
oli
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[David Barak]

--- Lawrence Wong <***@yahoo.com> wrote:
<questions regarding MLPPP>

Cisco MLPPP mostly works, but there are a couple of
really bad gotchas. My experience with it is watching
it NOT work on 75xx series routers. More accurately,
it would work, but when a T1 would bounce and recover,
a show command would cause the router to reload. That
was not resolved as of last October (when I left the
company which had the case open).

My overall impression is that MLPPP is relatively
flaky. I could recommend MLFR for point-to-point
circuits, but that can be kind of flaky too (lots of
problems in initial code releases), although I've now
got some customers using this, and it's working most
of the time.

Do you own both ends of the circuit?

-David Barak
need Geek Rock? Try The Franchise!
http://www.listentothefranchise.com



__________________________________
Do you Yahoo!?
Make Yahoo! your home page
http://www.yahoo.com/r/hs
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Rodney Dunn]

I've worked through a lot of those bugs on the 75xx
and we did have some bad ones.

All the ones I know of have been fixed in later 12.0S
code and latest 12.2 and 12.3 mainline
code.

You were most likely hitting:

CSCed29590
Multilink PPP link flap causes output frozen for member link

which was the worst one.

I've seen lots of large customers running dMLP on
a 75xx running stable for over a year.

My recommendation would be 12.0(27)S4 or 12.3(13).

What I find is most people don't have the box setup
properly to be running in fully distributed mode so the
bundles are distributed and/or their VIPs don't have the
CPU power to handle the bundles.


Rodney


CSCeg28064
Internally found moderate defect: Assigned (R)
cbus complex results in incorrect multilink state

CSCeb31029
Internally found moderate defect: Resolved (R)
Output rates incorrect on dMLP/dMFR/dLFI interfaces under congestion

CSCee69493
Externally found moderate defect: Resolved (R)
dMLP fails if control is given to IOS MLPPP

CSCin36465 Watchdog crashed because of MLPPP

CSCin47712 txacc loss after OIR

CSCec00268 Input drops and * throttles on PPP multilink interface

CSCea59948 Output stuck on a T3 Port Adaptor

CSCec87815
Multilink buffer header leak on a VIP with an ATM PA installed

CSCed29590
Multilink PPP link flap causes output frozen for member link

CSCin55432
Internally found moderate defect: Resolved (R)
Spurious memory access at vipmlp_change_primary

Post by David Barak
<questions regarding MLPPP>
Cisco MLPPP mostly works, but there are a couple of
really bad gotchas. My experience with it is watching
it NOT work on 75xx series routers. More accurately,
it would work, but when a T1 would bounce and recover,
a show command would cause the router to reload. That
was not resolved as of last October (when I left the
company which had the case open).
My overall impression is that MLPPP is relatively
flaky. I could recommend MLFR for point-to-point
circuits, but that can be kind of flaky too (lots of
problems in initial code releases), although I've now
got some customers using this, and it's working most
of the time.
Do you own both ends of the circuit?
-David Barak
need Geek Rock? Try The Franchise!
http://www.listentothefranchise.com
__________________________________
Do you Yahoo!?
Make Yahoo! your home page
http://www.yahoo.com/r/hs
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Richard J. Sears]

Hi Rodney,

Would 7513's running 12.2(27) have problems with MLPPP...? We had to
convert all of our MLPPP customers to ip load sharing per packet because
we would lose the mlppp links and the only way to get them back was to
reboot the router. I would like to use MLPPP, but I (like many) have had
horrible headaches with it.

We are running around 300 T1s per 7513 with RSP4s.

Thanks

On Thu, 24 Mar 2005 09:24:39 -0500

Post by Rodney Dunn
I've worked through a lot of those bugs on the 75xx
and we did have some bad ones.
All the ones I know of have been fixed in later 12.0S
code and latest 12.2 and 12.3 mainline
code.
CSCed29590
Multilink PPP link flap causes output frozen for member link
which was the worst one.
I've seen lots of large customers running dMLP on
a 75xx running stable for over a year.
My recommendation would be 12.0(27)S4 or 12.3(13).
What I find is most people don't have the box setup
properly to be running in fully distributed mode so the
bundles are distributed and/or their VIPs don't have the
CPU power to handle the bundles.
Rodney
CSCeg28064
Internally found moderate defect: Assigned (R)
cbus complex results in incorrect multilink state
CSCeb31029
Internally found moderate defect: Resolved (R)
Output rates incorrect on dMLP/dMFR/dLFI interfaces under congestion
CSCee69493
Externally found moderate defect: Resolved (R)
dMLP fails if control is given to IOS MLPPP
CSCin36465 Watchdog crashed because of MLPPP
CSCin47712 txacc loss after OIR
CSCec00268 Input drops and * throttles on PPP multilink interface
CSCea59948 Output stuck on a T3 Port Adaptor
CSCec87815
Multilink buffer header leak on a VIP with an ATM PA installed
CSCed29590
Multilink PPP link flap causes output frozen for member link
CSCin55432
Internally found moderate defect: Resolved (R)
Spurious memory access at vipmlp_change_primary

Post by David Barak
<questions regarding MLPPP>
Cisco MLPPP mostly works, but there are a couple of
really bad gotchas. My experience with it is watching
it NOT work on 75xx series routers. More accurately,
it would work, but when a T1 would bounce and recover,
a show command would cause the router to reload. That
was not resolved as of last October (when I left the
company which had the case open).
My overall impression is that MLPPP is relatively
flaky. I could recommend MLFR for point-to-point
circuits, but that can be kind of flaky too (lots of
problems in initial code releases), although I've now
got some customers using this, and it's working most
of the time.
Do you own both ends of the circuit?
-David Barak
need Geek Rock? Try The Franchise!
http://www.listentothefranchise.com
__________________________________
Do you Yahoo!?
Make Yahoo! your home page
http://www.yahoo.com/r/hs
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

******************************************
Richard J. Sears
Vice President
American Internet Services
----------------------------------------------------
***@adnc.com
http://www.adnc.com
----------------------------------------------------
858.576.4272 - Phone
858.427.2401 - Fax
INOC-DBA - 6130
----------------------------------------------------

I fly because it releases my mind
from the tyranny of petty things . .


"Work like you don't need the money, love like you've
never been hurt and dance like you do when nobody's
watching."

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Chris Cappuccio]

Post by Richard J. Sears
Would 7513's running 12.2(27) have problems with MLPPP...? We had to
convert all of our MLPPP customers to ip load sharing per packet because
we would lose the mlppp links and the only way to get them back was to
reboot the router. I would like to use MLPPP, but I (like many) have had

Shit, I tried 12.2(18)S8 and it broke all my MLPPP connections that worked
fine on 12.2(18)S4. The situation is 7507/RSP4....everything was up/up upon
moving to S8, but no traffic passed, and ppp debug showed continuous "PROTREJ"
messages, as in, something is really broken. TAC is being slow to respond :)

[Rodney Dunn]

I heard there is a dCEF problem in that throttle
and they are working to defer it for the 75xx.

If you want MLPPP in 12.2S you had better wait a bit
until the new 12.2S image comes out that integrates
the MLPPP and MPLS stuff because a lot of MLPPP bugs
are fixed there.

For now, go with 12.3 or 12.0S.


Rodney

Post by Chris Cappuccio

Post by Richard J. Sears
Would 7513's running 12.2(27) have problems with MLPPP...? We had to
convert all of our MLPPP customers to ip load sharing per packet because
we would lose the mlppp links and the only way to get them back was to
reboot the router. I would like to use MLPPP, but I (like many) have had

Shit, I tried 12.2(18)S8 and it broke all my MLPPP connections that worked
fine on 12.2(18)S4. The situation is 7507/RSP4....everything was up/up upon
moving to S8, but no traffic passed, and ppp debug showed continuous "PROTREJ"
messages, as in, something is really broken. TAC is being slow to respond :)
--
"All creativity is an extended form of a joke. Most creativity is a transition
from one context into another where things are more surprising." - Alan Kay
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Chris Cappuccio]

Post by Rodney Dunn
I heard there is a dCEF problem in that throttle
and they are working to defer it for the 75xx.
If you want MLPPP in 12.2S you had better wait a bit
until the new 12.2S image comes out that integrates
the MLPPP and MPLS stuff because a lot of MLPPP bugs
are fixed there.
For now, go with 12.3 or 12.0S.

Well, it runs solid under 12.2(18)S4, for at least the last year....
It was only until I upgraded to S8 that I had the problem!
Downgrade back to S4, no problem.

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Rodney Dunn]

Yeah...I'm working to figure out what happened and
take action....

Post by Chris Cappuccio

Post by Rodney Dunn
I heard there is a dCEF problem in that throttle
and they are working to defer it for the 75xx.
If you want MLPPP in 12.2S you had better wait a bit
until the new 12.2S image comes out that integrates
the MLPPP and MPLS stuff because a lot of MLPPP bugs
are fixed there.
For now, go with 12.3 or 12.0S.

Well, it runs solid under 12.2(18)S4, for at least the last year....
It was only until I upgraded to S8 that I had the problem!
Downgrade back to S4, no problem.

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

Am I correct to say that if using per packet load
sharing, the maximum throughput at any one time will
be the max of the individual link and not the bundle
as a whole?

Post by Richard J. Sears
Hi Rodney,
Would 7513's running 12.2(27) have problems with
MLPPP...? We had to
convert all of our MLPPP customers to ip load
sharing per packet because
we would lose the mlppp links and the only way to
get them back was to
reboot the router. I would like to use MLPPP, but I
(like many) have had
horrible headaches with it.
We are running around 300 T1s per 7513 with RSP4s.
Thanks
On Thu, 24 Mar 2005 09:24:39 -0500

Post by Rodney Dunn
I've worked through a lot of those bugs on the

75xx

Post by Rodney Dunn
and we did have some bad ones.
All the ones I know of have been fixed in later

12.0S

Post by Rodney Dunn
code and latest 12.2 and 12.3 mainline
code.
CSCed29590
Multilink PPP link flap causes output frozen for

member link

Post by Rodney Dunn
which was the worst one.
I've seen lots of large customers running dMLP on
a 75xx running stable for over a year.
My recommendation would be 12.0(27)S4 or 12.3(13).
What I find is most people don't have the box

setup

Post by Rodney Dunn
properly to be running in fully distributed mode

so the

Post by Rodney Dunn
bundles are distributed and/or their VIPs don't

have the

Post by Rodney Dunn
CPU power to handle the bundles.
Rodney
CSCeg28064
Internally found moderate defect: Assigned (R)
cbus complex results in incorrect multilink state
CSCeb31029
Internally found moderate defect: Resolved (R)
Output rates incorrect on dMLP/dMFR/dLFI

interfaces under congestion

Post by Rodney Dunn
CSCee69493
Externally found moderate defect: Resolved (R)
dMLP fails if control is given to IOS MLPPP
CSCin36465 Watchdog crashed because of MLPPP
CSCin47712 txacc loss after OIR
CSCec00268 Input drops and * throttles on PPP

multilink interface

Post by Rodney Dunn
CSCea59948 Output stuck on a T3 Port Adaptor
CSCec87815
Multilink buffer header leak on a VIP with an ATM

PA installed

Post by Rodney Dunn
CSCed29590
Multilink PPP link flap causes output frozen for

member link

Post by Rodney Dunn
CSCin55432
Internally found moderate defect: Resolved (R)
Spurious memory access at vipmlp_change_primary
On Thu, Mar 24, 2005 at 06:01:48AM -0800, David

Post by David Barak
<questions regarding MLPPP>
Cisco MLPPP mostly works, but there are a couple

of

Post by Rodney Dunn

Post by David Barak
really bad gotchas. My experience with it is

watching

Post by Rodney Dunn

Post by David Barak
it NOT work on 75xx series routers. More

accurately,

Post by Rodney Dunn

Post by David Barak
it would work, but when a T1 would bounce and

recover,

Post by Rodney Dunn

Post by David Barak
a show command would cause the router to reload.

That

Post by Rodney Dunn

Post by David Barak
was not resolved as of last October (when I left

the

Post by Rodney Dunn

Post by David Barak
company which had the case open).
My overall impression is that MLPPP is

relatively

Post by Rodney Dunn

Post by David Barak
flaky. I could recommend MLFR for

point-to-point

Post by Rodney Dunn

Post by David Barak
circuits, but that can be kind of flaky too

(lots of

Post by Rodney Dunn

Post by David Barak
problems in initial code releases), although

I've now

Post by Rodney Dunn

Post by David Barak
got some customers using this, and it's working

most

Post by Rodney Dunn

Post by David Barak
of the time.
Do you own both ends of the circuit?
-David Barak
need Geek Rock? Try The Franchise!
http://www.listentothefranchise.com
__________________________________
Do you Yahoo!?
Make Yahoo! your home page
http://www.yahoo.com/r/hs
_______________________________________________
cisco-nsp mailing list

https://puck.nether.net/mailman/listinfo/cisco-nsp

Post by Rodney Dunn

Post by David Barak
archive at

http://puck.nether.net/pipermail/cisco-nsp/

Post by Rodney Dunn
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at

http://puck.nether.net/pipermail/cisco-nsp/
******************************************
Richard J. Sears
Vice President
American Internet Services
----------------------------------------------------
http://www.adnc.com
----------------------------------------------------
858.576.4272 - Phone
858.427.2401 - Fax
INOC-DBA - 6130
----------------------------------------------------
I fly because it releases my mind
from the tyranny of petty things . .
"Work like you don't need the money, love like
you've
never been hurt and dance like you do when nobody's
watching."
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at
http://puck.nether.net/pipermail/cisco-nsp/

__________________________________
Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Jon Lewis]

Post by Lawrence Wong
Am I correct to say that if using per packet load
sharing, the maximum throughput at any one time will
be the max of the individual link and not the bundle
as a whole?

No. That would be the default, per destination. Per packet lets any one
session utilize as much of the aggregate bandwidth as the endpoint systems
are able...limited by latency and window sizes.

----------------------------------------------------------------------
Jon Lewis | I route
Senior Network Engineer | therefore you are
Atlantic Net |
_________ http://www.lewis.org/~jlewis/pgp for PGP public key_________
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Christopher E. Brown]

Post by Lawrence Wong
Am I correct to say that if using per packet load
sharing, the maximum throughput at any one time will
be the max of the individual link and not the bundle
as a whole?

You have it inverted.

per-packet shortcuts CEF (does a lookup for each packet, spreads across
pipes)

per-flow (the default) uses CEF normally (one flow, one pipe)
--
------------------------------------------------------------------------
Christopher E. Brown <***@acsalaska.net> desk [phone]
cell [phone]
IP Engineer - ACS
------------------------------------------------------------------------
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Rodney Dunn]

Post by Richard J. Sears
Hi Rodney,
Would 7513's running 12.2(27) have problems with MLPPP...? We had to
convert all of our MLPPP customers to ip load sharing per packet because
we would lose the mlppp links and the only way to get them back was to
reboot the router. I would like to use MLPPP, but I (like many) have had
horrible headaches with it.
We are running around 300 T1s per 7513 with RSP4s.

The customer I worked with closely had 800+ interfaces
running at various rates with Gig uplinks to the core.

Leased line aggregation. Average uplink bw usage was
around 100 Mbps spread over two GEIP+ boards.

The downstreams are all MCT3 PA's channelized out.

No heavy features (QOS, ACL's etc..)

The maximum number of interfaces you can have is like
832 I think (that's the QA ASIC limiation).
That isn't software IDB's (tunnels, subints, etc..).


While I'd like to say all those bugs were back ported
to 12.2 I don't think that was done because at the time
we didn't have any customers doing it.

Your best bet would be 12.3 or 12.0S.

There will be a 12.2S version coming out for the 75xx
at some point that has more MPLS/MLPPP feature interaction
support.

Rodney

Post by Richard J. Sears
Thanks
On Thu, 24 Mar 2005 09:24:39 -0500

Post by Rodney Dunn
I've worked through a lot of those bugs on the 75xx
and we did have some bad ones.
All the ones I know of have been fixed in later 12.0S
code and latest 12.2 and 12.3 mainline
code.
CSCed29590
Multilink PPP link flap causes output frozen for member link
which was the worst one.
I've seen lots of large customers running dMLP on
a 75xx running stable for over a year.
My recommendation would be 12.0(27)S4 or 12.3(13).
What I find is most people don't have the box setup
properly to be running in fully distributed mode so the
bundles are distributed and/or their VIPs don't have the
CPU power to handle the bundles.
Rodney
CSCeg28064
Internally found moderate defect: Assigned (R)
cbus complex results in incorrect multilink state
CSCeb31029
Internally found moderate defect: Resolved (R)
Output rates incorrect on dMLP/dMFR/dLFI interfaces under congestion
CSCee69493
Externally found moderate defect: Resolved (R)
dMLP fails if control is given to IOS MLPPP
CSCin36465 Watchdog crashed because of MLPPP
CSCin47712 txacc loss after OIR
CSCec00268 Input drops and * throttles on PPP multilink interface
CSCea59948 Output stuck on a T3 Port Adaptor
CSCec87815
Multilink buffer header leak on a VIP with an ATM PA installed
CSCed29590
Multilink PPP link flap causes output frozen for member link
CSCin55432
Internally found moderate defect: Resolved (R)
Spurious memory access at vipmlp_change_primary

Post by David Barak
<questions regarding MLPPP>
Cisco MLPPP mostly works, but there are a couple of
really bad gotchas. My experience with it is watching
it NOT work on 75xx series routers. More accurately,
it would work, but when a T1 would bounce and recover,
a show command would cause the router to reload. That
was not resolved as of last October (when I left the
company which had the case open).
My overall impression is that MLPPP is relatively
flaky. I could recommend MLFR for point-to-point
circuits, but that can be kind of flaky too (lots of
problems in initial code releases), although I've now
got some customers using this, and it's working most
of the time.
Do you own both ends of the circuit?
-David Barak
need Geek Rock? Try The Franchise!
http://www.listentothefranchise.com
__________________________________
Do you Yahoo!?
Make Yahoo! your home page
http://www.yahoo.com/r/hs
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

******************************************
Richard J. Sears
Vice President
American Internet Services
----------------------------------------------------
http://www.adnc.com
----------------------------------------------------
858.576.4272 - Phone
858.427.2401 - Fax
INOC-DBA - 6130
----------------------------------------------------
I fly because it releases my mind
from the tyranny of petty things . .
"Work like you don't need the money, love like you've
never been hurt and dance like you do when nobody's
watching."

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[David Barak]

Post by Rodney Dunn
I've worked through a lot of those bugs on the 75xx
and we did have some bad ones.
All the ones I know of have been fixed in later
12.0S
code and latest 12.2 and 12.3 mainline
code.

Glad to hear that those have finally been addressed -
it was a source of much consternation (because our
competition for the client's business was using
another vendor's routers, and they loved to beat us up
about it).

Your note that you've had stable customers for a year
is very encouraging :)

-David Barak
need Geek Rock? Try The Franchise!
http://www.listentothefranchise.com





__________________________________
Do you Yahoo!?
Yahoo! Mail - 250MB free storage. Do more. Manage less.
http://info.mail.yahoo.com/mail_250
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Rodney Dunn]

I've got a few bruises to show for it still. :)

Post by David Barak

Post by Rodney Dunn
I've worked through a lot of those bugs on the 75xx
and we did have some bad ones.
All the ones I know of have been fixed in later
12.0S
code and latest 12.2 and 12.3 mainline
code.

Glad to hear that those have finally been addressed -
it was a source of much consternation (because our
competition for the client's business was using
another vendor's routers, and they loved to beat us up
about it).
Your note that you've had stable customers for a year
is very encouraging :)
-David Barak
need Geek Rock? Try The Franchise!
http://www.listentothefranchise.com
__________________________________
Do you Yahoo!?
Yahoo! Mail - 250MB free storage. Do more. Manage less.
http://info.mail.yahoo.com/mail_250
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

Post by Oliver Boehmer (oboehmer)
It is supported on all low- to mid-end routers up to
and including 7500,
check the release notes and/or feature navigator for
higher-end as well
as hardware-based platforms

Do you have any idea how well it will work on
"ancient" hardware like the 2514 and 2621? Cos
currently the E1 (sorry for the earlier typo of T1),
which connects my warehouse to my head office is

Post by Oliver Boehmer (oboehmer)

Post by Lawrence Wong
Under MLPPP, will traffic be distributed amongst

both

Post by Lawrence Wong
links (i.e. I can download at max of 4Mbps) or am

I

Post by Lawrence Wong
limited to the max of each link (i.e. I can

download

Post by Lawrence Wong
at max of 2Mbps).

A single session can use up to 4Mbps. MLPPP will
distribute the packets
over both links making sure to preserve the packet
sequence.

That's cool. It beats running OSPF over multiple
links.

What's the lead time like to detect a link failure in
a MLPPP bundle?

Post by Oliver Boehmer (oboehmer)

Post by Lawrence Wong
How processor intensive is MLPPP?

In halfway recent code it is CEF switched and
2x2Mbps should be a
non-issue.. You can optimize by disabling
fragmentation, but it should
not be needed to do this.

Do you have any idea how scalable it can get for the
platform (2514 & 2621) metioned above? How many E1's
can I MLPPP before perform degrades? 3 x 2Mbps? 4 x
2Mbps?

Will using Cisco 1845 be a better choice for 3 or more
2Mbps MLPPP links? I am running just plain WAN
connectivity with no bells and whistles.

Post by Oliver Boehmer (oboehmer)

Post by Lawrence Wong
Are there any caveats to look out for?

you want to make sure the delay over both links is
somewhat equal.
Otherwise MLPPP could run out of buffers trying to
reassemble fragments
or preserving the packet order.

Thanks,




__________________________________
Do you Yahoo!?
Yahoo! Sports - Sign up for Fantasy Baseball.
http://baseball.fantasysports.yahoo.com/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Bruce Robertson]

Don't try a 2514; not enough CPU. It'll work great on a 2621. We've bundled
four links on a 2651 with no problems.

--
Bruce Robertson, President/CEO [phone]
Great Basin Internet Services, Inc. company-wide fax: [phone]
http://www.greatbasin.net my efax: [phone]

Post by Lawrence Wong

Post by Oliver Boehmer (oboehmer)
It is supported on all low- to mid-end routers up to
and including 7500,
check the release notes and/or feature navigator for
higher-end as well
as hardware-based platforms

Do you have any idea how well it will work on
"ancient" hardware like the 2514 and 2621? Cos
currently the E1 (sorry for the earlier typo of T1),
which connects my warehouse to my head office is

Post by Oliver Boehmer (oboehmer)

Post by Lawrence Wong
Under MLPPP, will traffic be distributed amongst

both

Post by Lawrence Wong
links (i.e. I can download at max of 4Mbps) or am

I

Post by Lawrence Wong
limited to the max of each link (i.e. I can

download

Post by Lawrence Wong
at max of 2Mbps).

A single session can use up to 4Mbps. MLPPP will
distribute the packets
over both links making sure to preserve the packet
sequence.

That's cool. It beats running OSPF over multiple
links.
What's the lead time like to detect a link failure in
a MLPPP bundle?

Post by Oliver Boehmer (oboehmer)

Post by Lawrence Wong
How processor intensive is MLPPP?

In halfway recent code it is CEF switched and
2x2Mbps should be a
non-issue.. You can optimize by disabling
fragmentation, but it should
not be needed to do this.

Do you have any idea how scalable it can get for the
platform (2514 & 2621) metioned above? How many E1's
can I MLPPP before perform degrades? 3 x 2Mbps? 4 x
2Mbps?
Will using Cisco 1845 be a better choice for 3 or more
2Mbps MLPPP links? I am running just plain WAN
connectivity with no bells and whistles.

Post by Oliver Boehmer (oboehmer)

Post by Lawrence Wong
Are there any caveats to look out for?

you want to make sure the delay over both links is
somewhat equal.
Otherwise MLPPP could run out of buffers trying to
reassemble fragments
or preserving the packet order.

Thanks,
__________________________________
Do you Yahoo!?
Yahoo! Sports - Sign up for Fantasy Baseball.
http://baseball.fantasysports.yahoo.com/
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Bruce Robertson]

Post by Lawrence Wong
What's the lead time like to detect a link failure in
a MLPPP bundle?

Once the link failure is detected, it's pretty much instantaneous. If you
test it with 'shutdown', though, you'll see a delay of a few seconds unless
you configure 'keepalive 1' on the individual T1s.

--
Bruce Robertson, President/CEO [phone]
Great Basin Internet Services, Inc. company-wide fax: [phone]
http://www.greatbasin.net my efax: [phone]
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Rodney Dunn]

Right on the money.

There are two driving forces.

a) L1 fault detection that brings the line down
b) L2 PPP keepalive timers set on the *member* links
of the bundle. Minimum time is 1 sec.
I haven't tried setting it to 1sec with no retry.
That seems pretty dangerous for false positives to me.

Given that L1 fault detection can be on the order of
2 seconds I'd say 1 sec keeps on the member links with
3 retries would be as low as I'd go.

Rodney

Post by Bruce Robertson

Post by Lawrence Wong
What's the lead time like to detect a link failure in
a MLPPP bundle?

Once the link failure is detected, it's pretty much instantaneous. If you
test it with 'shutdown', though, you'll see a delay of a few seconds unless
you configure 'keepalive 1' on the individual T1s.
--
Bruce Robertson, President/CEO [phone]
Great Basin Internet Services, Inc. company-wide fax: [phone]
http://www.greatbasin.net my efax: [phone]
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Jon Lewis]

Post by Lawrence Wong
Are there any caveats to look out for?

The one that still bugs the heck out of me is that almost any change to a
multilink interface (like application or removal of QoS settings) causes
the interface to flap.

I just converted one of our remote POPs from an NxT1 cef load sharing to
MLPPP to see if it would help with some VOIP quality issues we were
having. Am I right in thinking that MLPPP avoids the sort of out of order
packet arrival that per-packet cef load sharing can cause?...while still
doing nearly perfect load sharing across a number of T1s?

----------------------------------------------------------------------
Jon Lewis | I route
Senior Network Engineer | therefore you are
Atlantic Net |
_________ http://www.lewis.org/~jlewis/pgp for PGP public key_________
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

Hi,

Can you share how you run multilink PPP on a Cisco
831? I was under the impression that the 831 has only
1 WAN interface.

Best regards,

Post by Mohacsi Janos
Except Cisco 8xx. We have serious problems with
Cisco 831 with multilink
PPP....
Regards,
Janos Mohacsi
Network Engineer, Research Associate
NIIF/HUNGARNET, HUNGARY
Key 00F9AF98: 8645 1312 D249 471B DBAE 21A2 9F52
0D1F 00F9 AF98

__________________________________
Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

Is there any implication for setting "too low" a
keepalive?

Post by Lawrence Wong

Post by Lawrence Wong
What's the lead time like to detect a link

failure in

Post by Lawrence Wong
a MLPPP bundle?

Once the link failure is detected, it's pretty much
instantaneous. If you
test it with 'shutdown', though, you'll see a delay
of a few seconds unless
you configure 'keepalive 1' on the individual T1s.
--
Bruce Robertson, President/CEO
[phone]
Great Basin Internet Services, Inc. company-wide
fax: [phone]
http://www.greatbasin.net my
efax: [phone]

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Rodney Dunn]

The keepalives are supposed to be protected
against congestion on transmit so the links
shouldn't flap on congestion due to L2 keepalive
loss.

However, given the L1 fault detection times on
the circuit (least common denominator theory)
I don't see any reason to set them lower than
2-3 seconds for retry.

I usually recommend 1 and 3.

Rodney

Post by Lawrence Wong
Is there any implication for setting "too low" a
keepalive?

Post by Lawrence Wong

Post by Lawrence Wong
What's the lead time like to detect a link

failure in

Post by Lawrence Wong
a MLPPP bundle?

Once the link failure is detected, it's pretty much
instantaneous. If you
test it with 'shutdown', though, you'll see a delay
of a few seconds unless
you configure 'keepalive 1' on the individual T1s.
--
Bruce Robertson, President/CEO
[phone]
Great Basin Internet Services, Inc. company-wide
fax: [phone]
http://www.greatbasin.net my
efax: [phone]

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Oliver Boehmer (oboehmer)]

Post by Jon Lewis
I just converted one of our remote POPs from an NxT1 cef load sharing
to MLPPP to see if it would help with some VOIP quality issues we were
having. Am I right in thinking that MLPPP avoids the sort of out of
order packet arrival that per-packet cef load sharing can
cause?...while still doing nearly perfect load sharing across a
number of T1s?

Yes, both true..

oli

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Ted Mittelstaedt]

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
I just converted one of our remote POPs from an NxT1 cef load sharing
to MLPPP to see if it would help with some VOIP quality issues we
were having. Am I right in thinking that MLPPP avoids the sort of
out of order packet arrival that per-packet cef load sharing can
cause?...while still doing nearly perfect load sharing across a
number of T1s?

Yes, both true..

FALSE!

It does great load sharing but you can still have packets arriving out
of order.

I just closed a case with TAC on this last week. It was one of those
cases pushed up from bonehead support to higher level support. The
problem is CBAC ACL's didn't work on routers running load balancing
on either ip cef or multilink ppp. (well they worked if I only wanted
to have about 30kbps throughput on a 3MBps link, that is)

The tech the case eventually arrived at said in no uncertain terms that
this was a known bug - CBAC can't handle out of order packets, and you
get
them in either load balancing scenario.

He said the fix is in the most current 12.3, and that there have been
"literally thousands" of bugs corrected in load balancing scenarios in
the last year. (his words)

Ted

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Dennis Peng]

Post by Ted Mittelstaedt

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
I just converted one of our remote POPs from an NxT1 cef load sharing
to MLPPP to see if it would help with some VOIP quality issues we
were having. Am I right in thinking that MLPPP avoids the sort of
out of order packet arrival that per-packet cef load sharing can
cause?...while still doing nearly perfect load sharing across a
number of T1s?

Yes, both true..

FALSE!
It does great load sharing but you can still have packets arriving out
of order.

Can you explain how this occurs? Packets should never be sent
out-of-order through a multilink bundle.

Could you also point me to your case number? Thanks.

Dennis

Post by Ted Mittelstaedt
I just closed a case with TAC on this last week. It was one of those
cases pushed up from bonehead support to higher level support. The
problem is CBAC ACL's didn't work on routers running load balancing
on either ip cef or multilink ppp. (well they worked if I only wanted
to have about 30kbps throughput on a 3MBps link, that is)
The tech the case eventually arrived at said in no uncertain terms that
this was a known bug - CBAC can't handle out of order packets, and you
get
them in either load balancing scenario.
He said the fix is in the most current 12.3, and that there have been
"literally thousands" of bugs corrected in load balancing scenarios in
the last year. (his words)
Ted
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Jon Lewis]

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
were having. Am I right in thinking that MLPPP avoids the sort of
out of order packet arrival that per-packet cef load sharing can
cause?...while still doing nearly perfect load sharing across a
number of T1s?

Yes, both true..

With framentation disabled and Nx parrallel/identical T1 circuits in the
multilink bundle, are those assumptions (really just the ordering) still
[supposed to be] true?

For the POP I converted, which is now a 4 T1 multilink group, I'm seeing
as close to perfect load sharing as can be measured.

----------------------------------------------------------------------
Jon Lewis | I route
Senior Network Engineer | therefore you are
Atlantic Net |
_________ http://www.lewis.org/~jlewis/pgp for PGP public key_________
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Dennis Peng]

Post by Jon Lewis

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
were having. Am I right in thinking that MLPPP avoids the sort of
out of order packet arrival that per-packet cef load sharing can
cause?...while still doing nearly perfect load sharing across a
number of T1s?

Yes, both true..

With framentation disabled and Nx parrallel/identical T1 circuits in the
multilink bundle, are those assumptions (really just the ordering) still
[supposed to be] true?

Yes, even with fragmentation disabled, the packet is still sent with
the multilink headers which contain the sequence number. This allows
us to maintain the ordering of the packets.

Dennis

Post by Jon Lewis
For the POP I converted, which is now a 4 T1 multilink group, I'm seeing
as close to perfect load sharing as can be measured.
----------------------------------------------------------------------
Jon Lewis | I route
Senior Network Engineer | therefore you are
Atlantic Net |
_________ http://www.lewis.org/~jlewis/pgp for PGP public key_________

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Ted Mittelstaedt]

Post by Dennis Peng

Post by Ted Mittelstaedt

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
I just converted one of our remote POPs from an NxT1 cef load
sharing to MLPPP to see if it would help with some VOIP quality
issues we were having. Am I right in thinking that MLPPP avoids
the sort of out of order packet arrival that per-packet cef load
sharing can cause?...while still doing nearly perfect load sharing
across a number of T1s?

Yes, both true..

FALSE!
It does great load sharing but you can still have packets arriving
out of order.

Can you explain how this occurs? Packets should never be sent
out-of-order through a multilink bundle.

I am not going to attempt to explain it - you can query your coworker
Michael Wall at Cisco who told me this. I told him repeatedly that it
failed on both ip cef and multilink and he insisted it was an
out-of-order
issue. If he is full of shit, then educate him,
re-open the TAC case, and get to work fixing the bug.

Post by Dennis Peng
Could you also point me to your case number? Thanks.

601038235 Titled "FW Feature and dual T1s IOS don't work together"

Ted

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Cheung, Rick]

I've noticed it takes up to a minute for a 7200 at 12.3.12 with
a PA-MC-2T3+ to detect an unplugged T1 circuit on a 2620 at 12.3.12.10,
with 2 point to point T1s in a multilink bundle.

I've been reluctant to configure inband management, as we've
seen issues with frame relay end to end keepalives that time out upon
network load. The timers were hello intervals of 1 second, over a 30
second window. Different technology, I know.

I'll try the keep 1 3 tonight and see if it improves when
configured on the four serial interfaces.



Thanks,
Rick Cheung

-----Original Message-----
From: cisco-nsp-***@puck.nether.net
[mailto:cisco-nsp-***@puck.nether.net] On Behalf Of Rodney Dunn
Sent: Thursday, March 24, 2005 10:30 AM
To: Lawrence Wong
Cc: Bruce Robertson; cisco-***@puck.nether.net
Subject: Re: [c-nsp] Cisco MultiLink PPP

The keepalives are supposed to be protected
against congestion on transmit so the links
shouldn't flap on congestion due to L2 keepalive
loss.

However, given the L1 fault detection times on
the circuit (least common denominator theory)
I don't see any reason to set them lower than
2-3 seconds for retry.

I usually recommend 1 and 3.

Rodney

Post by Lawrence Wong
Is there any implication for setting "too low" a
keepalive?

Post by Lawrence Wong

Post by Lawrence Wong
What's the lead time like to detect a link

failure in

Post by Lawrence Wong
a MLPPP bundle?

Once the link failure is detected, it's pretty much
instantaneous. If you
test it with 'shutdown', though, you'll see a delay
of a few seconds unless
you configure 'keepalive 1' on the individual T1s.
--
Bruce Robertson, President/CEO
[phone]
Great Basin Internet Services, Inc. company-wide
fax: [phone]
http://www.greatbasin.net my
efax: [phone]

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

This message, including any attachments, contains confidential information intended for a specific
individual and purpose and is protected by law. If you are not the intended recipient, please contact
sender immediately by reply e-mail and destroy all copies.
You are hereby notified that any disclosure, copying, or distribution of this message, or the taking
of any action based on it, is strictly prohibited.

WARNING: Computer viruses can be transmitted via email. The recipient should check this email
and any attachments for the presence of viruses. The sender accepts no liability for any damage
caused by any virus transmitted by this email. E-mail transmission cannot be guaranteed
to be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive
late or incomplete, or contain viruses. The sender therefore does not accept liability for any errors
or omissions in the contents of this message, which arise as a result of e-mail transmission.

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Kinczli Zoltán]

good point :)

I'm aware of Janos's problem and it is not link aggregation related.
MLPPP is not only link 'bundling' tool but a QoS/fragmentation tool as well.

Janos's problem is with MLPPP fragmentation with PPPoE encapsulation (and QoS)...
Which is currently not yet supported, but we hope, it will,... soon?

So, the problem is not with the multilink PPP.

regards,
-zoltan

-----Original Message-----
From: Lawrence Wong [mailto:***@yahoo.com]
Sent: Thursday, March 24, 2005 4:02 PM
To: Mohacsi Janos; Oliver Boehmer (oboehmer)
Cc: cisco-***@puck.nether.net
Subject: RE: [c-nsp] Cisco MultiLink PPP


Hi,

Can you share how you run multilink PPP on a Cisco
831? I was under the impression that the 831 has only
1 WAN interface.

Best regards,

Post by Mohacsi Janos
Except Cisco 8xx. We have serious problems with
Cisco 831 with multilink
PPP....
Regards,
Janos Mohacsi
Network Engineer, Research Associate
NIIF/HUNGARNET, HUNGARY
Key 00F9AF98: 8645 1312 D249 471B DBAE 21A2 9F52
0D1F 00F9 AF98

__________________________________
Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/
--
No virus found in this incoming message.
Checked by AVG Anti-Virus.
Version: 7.0.308 / Virus Database: 266.8.1 - Release Date: 3/23/2005
--
No virus found in this outgoing message.
Checked by AVG Anti-Virus.
Version: 7.0.308 / Virus Database: 266.8.1 - Release Date: 3/23/2005


Ez az üzenet és a hozzá kapcsolódó fájlok, tervezetek kizárólag a
Címzettnek szólnak, a bennük foglalt információk bizalmasak, melyek
titokban maradásához a Synergon Informatika Rt.-nek jogilag méltányolható
érdeke fuzodik. Amennyiben valamely hiba folytán Ön nem a címzettje ennek a
levélnek, kérjük, semmisítse meg, és értesítse az üzenet küldojét. Az
üzenet az elküldés elott vírusellenorzésen esett át, de a vírusmentességére
nincs semmilyen garancia, ezért kérjük, ellenorizze azt!

DISCLAIMER

This e-mail and any attached files are confidential and may be legally
privileged. The content of this e-mail is subject of efforts by Synergon to
maintain its confidentiality. Also this e-mail is intended for the sole use
of the individual or entity to whom it is addressed. If you are not the
addressee, and received this transmission in error please delete this
e-mail and notify its sender immediately. This e-mail message has been
checked for computer viruses but it could still be infected. Please test it
for viruses before use.



_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Ted Mittelstaedt]

Post by Lawrence Wong
Hi all,
I am thinking of combining 2 x T1 (1.5Mbps) into a
larger pipe of 3Mbps for greater capacity via MLPPP.
I would like to check if MLPPP is supported on all
Cisco platforms with Serial interface or is it only
supported on selected platforms.
Under MLPPP, will traffic be distributed amongst both
links (i.e. I can download at max of 4Mbps) or am I
limited to the max of each link (i.e. I can download
at max of 2Mbps).
How processor intensive is MLPPP?
Are there any caveats to look out for?
Does anyone have experience with MLPPP to share?

We have the following setup running and it works great:

7206 IOS 12.2.27

3600 IOS 12.2.27

2 T1s point to point.

CPU utilization on the 7206 peaks at 25% but averages
around 7% And the 7206 has a DS3 card in it and a 4
port serial card for the T1s and does a lot of other stuff.

On the 3600 it peaks at 45% but averages 25%

Customer bandwidth peaks at around 1.5Mbt, except at night
when they run their offsite backup where it saturates the
pipe at 3Mbt.

Ted
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

Woops, looks like I have to give up the trusty old
2514 that has been working fine for the past 9 years.
:(

Any idea how many E1 MLPPP is supported on the 2621? I
don't think I will grow beyond 2 channels but if I
need to replace the 2514, it would be on the safe side
to cater for 3-4 E1 MLPPP just in case (would a 3620
be good enough?) .

I am thinking of using the VWIC-2MFT-G703 module on
the 2621 to connect the individual links instead of
getting external v.35 convertors. Anyone has
experience with it?

TIA!

Post by Bruce Robertson
Don't try a 2514; not enough CPU. It'll work great
on a 2621. We've bundled
four links on a 2651 with no problems.
--
Bruce Robertson, President/CEO
[phone]
Great Basin Internet Services, Inc. company-wide
fax: [phone]
http://www.greatbasin.net my
efax: [phone]

--- "Oliver Boehmer (oboehmer)"

Post by Oliver Boehmer (oboehmer)
It is supported on all low- to mid-end routers up

to

Post by Oliver Boehmer (oboehmer)
and including 7500,
check the release notes and/or feature navigator

for

Post by Oliver Boehmer (oboehmer)
higher-end as well
as hardware-based platforms

Do you have any idea how well it will work on
"ancient" hardware like the 2514 and 2621? Cos
currently the E1 (sorry for the earlier typo of

T1),

which connects my warehouse to my head office is

&

Post by Oliver Boehmer (oboehmer)

Post by Lawrence Wong
Under MLPPP, will traffic be distributed amongst

both

Post by Lawrence Wong
links (i.e. I can download at max of 4Mbps) or am

I

Post by Lawrence Wong
limited to the max of each link (i.e. I can

download

Post by Lawrence Wong
at max of 2Mbps).

A single session can use up to 4Mbps. MLPPP will
distribute the packets
over both links making sure to preserve the packet
sequence.

That's cool. It beats running OSPF over multiple
links.
What's the lead time like to detect a link failure

in

a MLPPP bundle?

Post by Oliver Boehmer (oboehmer)

Post by Lawrence Wong
How processor intensive is MLPPP?

In halfway recent code it is CEF switched and
2x2Mbps should be a
non-issue.. You can optimize by disabling
fragmentation, but it should
not be needed to do this.

Do you have any idea how scalable it can get for

the

platform (2514 & 2621) metioned above? How many

E1's

can I MLPPP before perform degrades? 3 x 2Mbps? 4

x

2Mbps?
Will using Cisco 1845 be a better choice for 3 or

more

2Mbps MLPPP links? I am running just plain WAN
connectivity with no bells and whistles.

Post by Oliver Boehmer (oboehmer)

Post by Lawrence Wong
Are there any caveats to look out for?

you want to make sure the delay over both links is
somewhat equal.
Otherwise MLPPP could run out of buffers trying to
reassemble fragments
or preserving the packet order.

Thanks,
__________________________________
Do you Yahoo!?
Yahoo! Sports - Sign up for Fantasy Baseball.
http://baseball.fantasysports.yahoo.com/
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at

http://puck.nether.net/pipermail/cisco-nsp/

__________________________________
Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Gert Doering]

Hi,

Post by Lawrence Wong
Any idea how many E1 MLPPP is supported on the 2621? I
don't think I will grow beyond 2 channels but if I
need to replace the 2514, it would be on the safe side
to cater for 3-4 E1 MLPPP just in case (would a 3620
be good enough?) .

Stay away from the 3620. It's incredibly noisy, quite slow, needs much
space in the rack (1RU, but very 'deep'), can't take useful amounts of
memory, has no onboard LAN ports, and is end-of-life.

A 2621XM would be a good choice (faster, more memory, and less footprint
than the 3620).

Post by Lawrence Wong
I am thinking of using the VWIC-2MFT-G703 module on
the 2621 to connect the individual links instead of
getting external v.35 convertors. Anyone has
experience with it?

We've equipped a 2611 with 2 VWIC-2MFT-E1 and run 4x E1 on it (not with
MPPP, though) and it just worked. Can't say what the peak CPU load was
exactly, but it was never on "worrisome" levels.

The 2621 is faster, so it should do 4x E1 MPPP.

gert

[Joe Maimon]

Post by Gert Doering
Hi,

Post by Lawrence Wong
Any idea how many E1 MLPPP is supported on the 2621? I
don't think I will grow beyond 2 channels but if I
need to replace the 2514, it would be on the safe side
to cater for 3-4 E1 MLPPP just in case (would a 3620
be good enough?) .

Stay away from the 3620. It's incredibly noisy, quite slow, needs much
space in the rack (1RU, but very 'deep'), can't take useful amounts of
memory, has no onboard LAN ports, and is end-of-life.
A 2621XM would be a good choice (faster, more memory, and less footprint
than the 3620).

3620 is cheaper resold than XM, supports 4 FE interfaces, with 4 WIC
slots with apropriate NM's, should have enough ram/flash to handle 12.4
when it gets here, not to mention all the 12.3 that the vanilla 26xx cant.

Which can translate easily enough into 3 100mbps FD PPPoE connections
(something that currently requires dedicated ethernet interfaces to be
done properly) and one dot1q lan interface with CBAC/ACL/NAT firewalling
plus 4 (or more) T's.

Toss in an AIM and you should be able to handle up to 30mbps of
encryption. Nothing to sneeze at.

So this is a nice enough platform if you are looking for something
better than resold 26xx - non XM.

Both the 2620 and 3620 are better than all 17xx due to lack of dot1q
support, which is something I tried to raise on this list previously.

Joe
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Gert Doering]

Hi,

Post by Joe Maimon

Post by Gert Doering
Stay away from the 3620. It's incredibly noisy, quite slow, needs much
space in the rack (1RU, but very 'deep'), can't take useful amounts of
memory, has no onboard LAN ports, and is end-of-life.
A 2621XM would be a good choice (faster, more memory, and less footprint
than the 3620).

I have to correct me here: the 2621XM can *not* take more memory than the
3620. The 2651XM can.

Post by Joe Maimon
3620 is cheaper resold than XM,

Yes... (and there is good reason)

Post by Joe Maimon
supports 4 FE interfaces,

... but as soon as you add 2 FEs, it is likely to be more expensive...

... and in addition, it makes no sense at all to put more than 1 FE into
a 3620. The box is just too slow.

Post by Joe Maimon
with 4 WIC
slots with apropriate NM's, should have enough ram/flash to handle 12.4
when it gets here, not to mention all the 12.3 that the vanilla 26xx cant.

Uh. I'm not sure which sort of 3620 you're talking about. Certainly not
the 3620s we own.

261x/262x can take the same amount of dram and flash than a 3620 can take
(64Mb DRAM, 32Mb Flash), and there *already* are features that are dropped
out of the 3620 code base (like "OSPFv3 for IPv6") because the 3620 is
maxed out.

I'd wager a bet that there will be no 12.4 for 3620...

Post by Joe Maimon
Which can translate easily enough into 3 100mbps FD PPPoE connections
(something that currently requires dedicated ethernet interfaces to be
done properly) and one dot1q lan interface with CBAC/ACL/NAT firewalling
plus 4 (or more) T's.
Toss in an AIM and you should be able to handle up to 30mbps of
encryption. Nothing to sneeze at.

I'm unsure whether a 3620 will be able to even carry 30 Mbit/s. of
*unencrypted* traffic - even if it's only LAN-to-LAN, with no PPPoE
overhead and no multilink. The AIM won't help you here.

If you're looking for something with lots of Ethernet interfaces and
nice crypto performance, get a 1712.

Post by Joe Maimon
So this is a nice enough platform if you are looking for something
better than resold 26xx - non XM.

Anything is better than a 3620 (well, maybe except a 25xx) - even a
good old 4700M is more powerful.

Post by Joe Maimon
Both the 2620 and 3620 are better than all 17xx due to lack of dot1q
support, which is something I tried to raise on this list previously.

Can't comment on that, never tried dot1q on any 17xx. It would surprise
me if the 1712 can't do it - after all, it has a WIC-4ESW built-in, which
can do dot1q on other platforms.

gert

[Chris Roberts]

Post by Joe Maimon

Post by Joe Maimon
Both the 2620 and 3620 are better than all 17xx due to lack

of dot1q

Post by Joe Maimon
support, which is something I tried to raise on this list

previously.
Can't comment on that, never tried dot1q on any 17xx. It
would surprise me if the 1712 can't do it - after all, it has
a WIC-4ESW built-in, which can do dot1q on other platforms.

1760 does dot1q, no problems.
...
cisco 1760 (MPC860P) processor (revision 0x500) with 54212K/11324K bytes of
memory.
Processor board ID FOC0823333S (453933451), with hardware revision 0000
MPC860P processor: part number 5, mask 2
...

interface FastEthernet0/0
no ip address
speed 100
full-duplex
!
interface FastEthernet0/0.1
description xxxx Server VLAN
encapsulation dot1Q 1 native
ip address x.x.1.240 255.255.255.0
!
interface FastEthernet0/0.2
description LES circuit to xxxx
encapsulation dot1Q 2
ip address x.x.99.1 255.255.255.252


Cheers,
Chris.

---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.859 / Virus Database: 585 - Release Date: 13/02/2005


_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Denis V. Schapov]

Post by Gert Doering

Post by Gert Doering
A 2621XM would be a good choice (faster, more memory, and less footprint
than the 3620).

I have to correct me here: the 2621XM can *not* take more memory than the
3620. The 2651XM can.

http://www.cisco.com/en/US/products/hw/routers/ps259/prod_bulletin0900aecd801e3d53.html

My priority list in increasing order for such price/performance category ;)

1) 2600 non XM
2) 3620 (sometime has a problem with fake OIR triggering)
3) 265x non XM (DRAM maximum 128MB)
4) 2600XM
5) 2801/2811
....

Both 1) and 2), 3) are EOS, EOL. For 1) and 2) IOS images/features were reduced beggining
12.2(8)T, 12.2(13)T to feet into available flash/DRAM. 1), 2) and 3) has no support of new
voice/data modules.
It's very likely that there will be no 12.4 support for 3620 and officially for 2600 _non_
XM, but 2600 non XM has a chance to run some of 12.4 images that feet into flash/DRAM if
Cisco will not specially remove them from the code.

One more thing for 16MB flash equipped 261x
http://www.cisco.com/warp/public/770/fn62030.shtml


Denis.


_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Andre Beck]

Hi,

Post by Lawrence Wong
Woops, looks like I have to give up the trusty old
2514 that has been working fine for the past 9 years.
:(

If you're going to build MP bundles on E1, surely.

Post by Lawrence Wong
Any idea how many E1 MLPPP is supported on the 2621? I

The 2621 is quasi-identical to the 2620. I have first hand experience
with trying to upgrade a 2xE1 MP bundle on such box to 3xE1. It did
work in principle, but the CPU load on this box reached peaks beyond
60% when that bundle got fully loaded with traffic. This caused an
unusual amount of RTT jitter and response loss on pings to that router
what finally convinced us to give up the MP and retry with CEF load
sharing per-packet. The box has 15 to 20% less CPU load this way.

However, keep in mind that this was with a late 12.1, not really a
recent IOS. As already pointed out by others here, newer IOSs can do
MP CEF switched, so the numbers might be better there. This is something
I'd actually want to lab test soon as I'd really like to get back to MP
there if possible.

Post by Lawrence Wong
don't think I will grow beyond 2 channels but if I

Two channels is working great. You get all the advantage of MP including
perfect load balancing, guaranteed no reordering and load balancing of
IP multicast (something I can't stop repeating you *don't* ever get with
CEF per-packet).

Post by Lawrence Wong
need to replace the 2514, it would be on the safe side
to cater for 3-4 E1 MLPPP just in case (would a 3620
be good enough?) .

Forget 3620. Rather consider the newer 26xx models or even 18xx/28xx. They
aren't even really expensive compared to former gear. If ever going 3k
series, rather consider 37xx or 38xx, with exception for the 3660.

Post by Lawrence Wong
I am thinking of using the VWIC-2MFT-G703 module on
the 2621 to connect the individual links instead of
getting external v.35 convertors.

Do you *really* need unframed E1? It's the only additional virtue of the
-G703 over the -E1 and surely not worth the significant price step except
you must have it. You can even stamp a G.704 frame onto a line that is
provided unframed, so the only reason you must have unframed is when
talking to a remote CSU/DSU that is limited to unframed either by design
or by not beeing under your control.

Post by Lawrence Wong
Anyone has experience with it?

So far only with the -E1. Work like a charm. We've got several 26xx
(including 2620) packed with those VWICs and they are doing great.
For proper MP operation with 2xE1 and CBAC we prefer the 2650 class,
but otherwise a 2620 can well deal with 4 individual E1 interfaces at
full load without a tendency to fall over.

HTH,
Andre.

[Lawrence Wong]

Is there any features that can run on a normal serial
interface that I can't run on a MLPPP interface (i.e
CBWFQ, LLQ, IP firewall, etc)? Or is everything
supported?

Thanks,

Post by Jon Lewis

Post by Jon Lewis
I just converted one of our remote POPs from an

NxT1 cef load sharing

Post by Jon Lewis
to MLPPP to see if it would help with some VOIP

quality issues we were

Post by Jon Lewis
having. Am I right in thinking that MLPPP avoids

the sort of out of

Post by Jon Lewis
order packet arrival that per-packet cef load

sharing can

Post by Jon Lewis
cause?...while still doing nearly perfect load

sharing across a

Post by Jon Lewis
number of T1s?

Yes, both true..
oli
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at
http://puck.nether.net/pipermail/cisco-nsp/

__________________________________
Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

On the topic of 12.0S, do you know if the access list
numbering feature will be ported back to 12.0S? One
main reason why we used 12.2S on our 7505 is because
the access list numbering feature allows us to edit
our ACLs on the fly if need be.

Post by Rodney Dunn
I've worked through a lot of those bugs on the 75xx
and we did have some bad ones.
All the ones I know of have been fixed in later
12.0S
code and latest 12.2 and 12.3 mainline
code.
CSCed29590
Multilink PPP link flap causes output frozen for
member link
which was the worst one.
I've seen lots of large customers running dMLP on
a 75xx running stable for over a year.
My recommendation would be 12.0(27)S4 or 12.3(13).
What I find is most people don't have the box setup
properly to be running in fully distributed mode so
the
bundles are distributed and/or their VIPs don't have
the
CPU power to handle the bundles.
Rodney
CSCeg28064
Internally found moderate defect: Assigned (R)
cbus complex results in incorrect multilink state
CSCeb31029
Internally found moderate defect: Resolved (R)
Output rates incorrect on dMLP/dMFR/dLFI interfaces
under congestion
CSCee69493
Externally found moderate defect: Resolved (R)
dMLP fails if control is given to IOS MLPPP
CSCin36465 Watchdog crashed because of MLPPP
CSCin47712 txacc loss after OIR
CSCec00268 Input drops and * throttles on PPP
multilink interface
CSCea59948 Output stuck on a T3 Port Adaptor
CSCec87815
Multilink buffer header leak on a VIP with an ATM PA
installed
CSCed29590
Multilink PPP link flap causes output frozen for
member link
CSCin55432
Internally found moderate defect: Resolved (R)
Spurious memory access at vipmlp_change_primary
On Thu, Mar 24, 2005 at 06:01:48AM -0800, David

Post by David Barak
<questions regarding MLPPP>
Cisco MLPPP mostly works, but there are a couple

of

Post by David Barak
really bad gotchas. My experience with it is

watching

Post by David Barak
it NOT work on 75xx series routers. More

accurately,

Post by David Barak
it would work, but when a T1 would bounce and

recover,

Post by David Barak
a show command would cause the router to reload.

That

Post by David Barak
was not resolved as of last October (when I left

the

Post by David Barak
company which had the case open).
My overall impression is that MLPPP is relatively
flaky. I could recommend MLFR for point-to-point
circuits, but that can be kind of flaky too (lots

of

Post by David Barak
problems in initial code releases), although I've

now

Post by David Barak
got some customers using this, and it's working

most

Post by David Barak
of the time.
Do you own both ends of the circuit?
-David Barak
need Geek Rock? Try The Franchise!
http://www.listentothefranchise.com
__________________________________
Do you Yahoo!?
Make Yahoo! your home page
http://www.yahoo.com/r/hs
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at

http://puck.nether.net/pipermail/cisco-nsp/

__________________________________
Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Rodney Dunn]

Don't know.

Post by Lawrence Wong
On the topic of 12.0S, do you know if the access list
numbering feature will be ported back to 12.0S? One
main reason why we used 12.2S on our 7505 is because
the access list numbering feature allows us to edit
our ACLs on the fly if need be.

Post by Rodney Dunn
I've worked through a lot of those bugs on the 75xx
and we did have some bad ones.
All the ones I know of have been fixed in later
12.0S
code and latest 12.2 and 12.3 mainline
code.
CSCed29590
Multilink PPP link flap causes output frozen for
member link
which was the worst one.
I've seen lots of large customers running dMLP on
a 75xx running stable for over a year.
My recommendation would be 12.0(27)S4 or 12.3(13).
What I find is most people don't have the box setup
properly to be running in fully distributed mode so
the
bundles are distributed and/or their VIPs don't have
the
CPU power to handle the bundles.
Rodney
CSCeg28064
Internally found moderate defect: Assigned (R)
cbus complex results in incorrect multilink state
CSCeb31029
Internally found moderate defect: Resolved (R)
Output rates incorrect on dMLP/dMFR/dLFI interfaces
under congestion
CSCee69493
Externally found moderate defect: Resolved (R)
dMLP fails if control is given to IOS MLPPP
CSCin36465 Watchdog crashed because of MLPPP
CSCin47712 txacc loss after OIR
CSCec00268 Input drops and * throttles on PPP
multilink interface
CSCea59948 Output stuck on a T3 Port Adaptor
CSCec87815
Multilink buffer header leak on a VIP with an ATM PA
installed
CSCed29590
Multilink PPP link flap causes output frozen for
member link
CSCin55432
Internally found moderate defect: Resolved (R)
Spurious memory access at vipmlp_change_primary
On Thu, Mar 24, 2005 at 06:01:48AM -0800, David

Post by David Barak
<questions regarding MLPPP>
Cisco MLPPP mostly works, but there are a couple

of

Post by David Barak
really bad gotchas. My experience with it is

watching

Post by David Barak
it NOT work on 75xx series routers. More

accurately,

Post by David Barak
it would work, but when a T1 would bounce and

recover,

Post by David Barak
a show command would cause the router to reload.

That

Post by David Barak
was not resolved as of last October (when I left

the

Post by David Barak
company which had the case open).
My overall impression is that MLPPP is relatively
flaky. I could recommend MLFR for point-to-point
circuits, but that can be kind of flaky too (lots

of

Post by David Barak
problems in initial code releases), although I've

now

Post by David Barak
got some customers using this, and it's working

most

Post by David Barak
of the time.
Do you own both ends of the circuit?
-David Barak
need Geek Rock? Try The Franchise!
http://www.listentothefranchise.com
__________________________________
Do you Yahoo!?
Make Yahoo! your home page
http://www.yahoo.com/r/hs
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at

http://puck.nether.net/pipermail/cisco-nsp/

__________________________________
Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

Which version will this new 12.2S image belong to?

On an offtopic issue, I was running 12.2(18)S on my
7505 facing the Internet just fine but anything from
12.2(18)S1 to S8 kills it outright with dbus timeout
errors but 12.2(25)S to S3 works okay.

Post by Rodney Dunn
I heard there is a dCEF problem in that throttle
and they are working to defer it for the 75xx.
If you want MLPPP in 12.2S you had better wait a bit
until the new 12.2S image comes out that integrates
the MLPPP and MPLS stuff because a lot of MLPPP bugs
are fixed there.
For now, go with 12.3 or 12.0S.
Rodney
On Thu, Mar 24, 2005 at 09:46:06AM -0800, Chris

Post by Chris Cappuccio

Post by Richard J. Sears
Would 7513's running 12.2(27) have problems with

MLPPP...? We had to

Post by Chris Cappuccio

Post by Richard J. Sears
convert all of our MLPPP customers to ip load

sharing per packet because

Post by Chris Cappuccio

Post by Richard J. Sears
we would lose the mlppp links and the only way

to get them back was to

Post by Chris Cappuccio

Post by Richard J. Sears
reboot the router. I would like to use MLPPP,

but I (like many) have had

Post by Chris Cappuccio
Shit, I tried 12.2(18)S8 and it broke all my MLPPP

connections that worked

Post by Chris Cappuccio
fine on 12.2(18)S4. The situation is

7507/RSP4....everything was up/up upon

Post by Chris Cappuccio
moving to S8, but no traffic passed, and ppp debug

showed continuous "PROTREJ"

Post by Chris Cappuccio
messages, as in, something is really broken. TAC

is being slow to respond :)

Post by Chris Cappuccio
--
"All creativity is an extended form of a joke.

Most creativity is a transition

Post by Chris Cappuccio
from one context into another where things are

more surprising." - Alan Kay

Post by Chris Cappuccio
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at

http://puck.nether.net/pipermail/cisco-nsp/
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at
http://puck.nether.net/pipermail/cisco-nsp/

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Jon Lewis]

Post by Lawrence Wong
Which version will this new 12.2S image belong to?
On an offtopic issue, I was running 12.2(18)S on my
7505 facing the Internet just fine but anything from
12.2(18)S1 to S8 kills it outright with dbus timeout
errors but 12.2(25)S to S3 works okay.

I've got 12.2(18)S5 and 12.2(18)S6 running relatively fine on 7507s and a
7513. IIRC, we had some dCEF issues with 12.2(18)S5 but no show stoppers.

----------------------------------------------------------------------
Jon Lewis | I route
Senior Network Engineer | therefore you are
Atlantic Net |
_________ http://www.lewis.org/~jlewis/pgp for PGP public key_________
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Rodney Dunn]

Post by Lawrence Wong
Which version will this new 12.2S image belong to?

The name hasn't been officially set yet.

Post by Lawrence Wong
On an offtopic issue, I was running 12.2(18)S on my
7505 facing the Internet just fine but anything from
12.2(18)S1 to S8 kills it outright with dbus timeout
errors but 12.2(25)S to S3 works okay.

We are deferring 12.2(18)S8 for the 75xx.

Don't load it if you are running dCEF.

The bug that is fixing it is:

CSCdz84963

Sorry for the problems it caused.

Rodney

Post by Lawrence Wong

Post by Rodney Dunn
I heard there is a dCEF problem in that throttle
and they are working to defer it for the 75xx.
If you want MLPPP in 12.2S you had better wait a bit
until the new 12.2S image comes out that integrates
the MLPPP and MPLS stuff because a lot of MLPPP bugs
are fixed there.
For now, go with 12.3 or 12.0S.
Rodney
On Thu, Mar 24, 2005 at 09:46:06AM -0800, Chris

Post by Chris Cappuccio

Post by Richard J. Sears
Would 7513's running 12.2(27) have problems with

MLPPP...? We had to

Post by Chris Cappuccio

Post by Richard J. Sears
convert all of our MLPPP customers to ip load

sharing per packet because

Post by Chris Cappuccio

Post by Richard J. Sears
we would lose the mlppp links and the only way

to get them back was to

Post by Chris Cappuccio

Post by Richard J. Sears
reboot the router. I would like to use MLPPP,

but I (like many) have had

Post by Chris Cappuccio
Shit, I tried 12.2(18)S8 and it broke all my MLPPP

connections that worked

Post by Chris Cappuccio
fine on 12.2(18)S4. The situation is

7507/RSP4....everything was up/up upon

Post by Chris Cappuccio
moving to S8, but no traffic passed, and ppp debug

showed continuous "PROTREJ"

Post by Chris Cappuccio
messages, as in, something is really broken. TAC

is being slow to respond :)

Post by Chris Cappuccio
--
"All creativity is an extended form of a joke.

Most creativity is a transition

Post by Chris Cappuccio
from one context into another where things are

more surprising." - Alan Kay

Post by Chris Cappuccio
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at

http://puck.nether.net/pipermail/cisco-nsp/
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at
http://puck.nether.net/pipermail/cisco-nsp/

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

May I know if that is a 3620 or 3640? Any issues in
scaling up to 4 x T1/E1 ?

Post by Ted Mittelstaedt
We have the following setup running and it works
7206 IOS 12.2.27
3600 IOS 12.2.27
2 T1s point to point.
CPU utilization on the 7206 peaks at 25% but
averages
around 7% And the 7206 has a DS3 card in it and a 4
port serial card for the T1s and does a lot of other
stuff.
On the 3600 it peaks at 45% but averages 25%
Customer bandwidth peaks at around 1.5Mbt, except at
night
when they run their offsite backup where it
saturates the
pipe at 3Mbt.
Ted

__________________________________
Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Ted Mittelstaedt]

May I know if that is a 3620 or 3640? Any issues in scaling up to 4 x
T1/E1 ?

It is a 3620.

I am pretty sure it could scale just fine if you were not doing NAT on it
(mine is). But as for the 2500 platform that definitely will not do 2
t1's pegged.
It will, however, do 2 T1's if the bandwidth is pretty asymmectrical
(mostly
incoming or mostly outgoing)

Ted

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

Post by Jon Lewis

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
were having. Am I right in thinking that

MLPPP avoids the sort of

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
out of order packet arrival that per-packet

cef load sharing can

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
cause?...while still doing nearly perfect

load sharing across a

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
number of T1s?

Yes, both true..

With framentation disabled and Nx
parrallel/identical T1 circuits in the
multilink bundle, are those assumptions (really just
the ordering) still
[supposed to be] true?
For the POP I converted, which is now a 4 T1
multilink group, I'm seeing
as close to perfect load sharing as can be measured.

Is it a requirement to have N identical links (i.e. 4
x T1) or can I have 1 x T1 + 1 x 512kbps?




__________________________________
Do you Yahoo!?
Yahoo! Mail - 250MB free storage. Do more. Manage less.
http://info.mail.yahoo.com/mail_250
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Eric Knudson]

As long as the correct interface bandwidth is configured, the correct
weight should be assigned to each member link, viewable with a 'show
ppp multi.'



On Fri, 25 Mar 2005 23:52:27 -0800 (PST), Lawrence Wong

Post by Lawrence Wong

Post by Jon Lewis

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
were having. Am I right in thinking that

MLPPP avoids the sort of

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
out of order packet arrival that per-packet

cef load sharing can

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
cause?...while still doing nearly perfect

load sharing across a

Post by Oliver Boehmer (oboehmer)

Post by Jon Lewis
number of T1s?

Yes, both true..

With framentation disabled and Nx
parrallel/identical T1 circuits in the
multilink bundle, are those assumptions (really just
the ordering) still
[supposed to be] true?
For the POP I converted, which is now a 4 T1
multilink group, I'm seeing
as close to perfect load sharing as can be measured.

Is it a requirement to have N identical links (i.e. 4
x T1) or can I have 1 x T1 + 1 x 512kbps?
__________________________________
Do you Yahoo!?
Yahoo! Mail - 250MB free storage. Do more. Manage less.
http://info.mail.yahoo.com/mail_250
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Josh Gentry]

Post by Lawrence Wong
Hi all,
I am thinking of combining 2 x T1 (1.5Mbps) into a
larger pipe of 3Mbps for greater capacity via MLPPP.

I wonder if you really need MLPPP? If you aren't worried about VOIP or something
similiar, you might consider just doing per packet load balancing with CEF. The
company I work for does this to combine 2 T1's into a 3Mbps pipe, and it works well.

http://www.joshgentry.com/cisco/cisco-load.html

Thanks,

Josh

[Lawrence Wong]

Thanks for the suggestion. I'm thinking of using MLPPP
to create a single virtual link that works like a big
fat single link instead of having packets run wild
across a few links. This could make management easier
in future?

On Thu, Mar 24, 2005 at 12:15:48AM -0800, Lawrence

Post by Lawrence Wong
Hi all,
I am thinking of combining 2 x T1 (1.5Mbps) into a
larger pipe of 3Mbps for greater capacity via

MLPPP.
I wonder if you really need MLPPP? If you aren't
worried about VOIP or something
similiar, you might consider just doing per packet
load balancing with CEF. The
company I work for does this to combine 2 T1's into
a 3Mbps pipe, and it works well.
http://www.joshgentry.com/cisco/cisco-load.html
Thanks,
Josh
--
Josh Gentry
Good Art Lives at http://www.jkdrummond.com/

__________________________________
Do you Yahoo!?
Yahoo! Sports - Sign up for Fantasy Baseball.
http://baseball.fantasysports.yahoo.com/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

Thanks for clarifying.

Post by Yury Yaroshevsky

Post by Lawrence Wong
Am I correct to say that if using per packet load
sharing, the maximum throughput at any one time

will

Post by Lawrence Wong
be the max of the individual link and not the

bundle

Post by Lawrence Wong
as a whole?

No. That would be the default, per destination.
Per packet lets any one
session utilize as much of the aggregate bandwidth
as the endpoint systems
are able...limited by latency and window sizes.

----------------------------------------------------------------------

Post by Yury Yaroshevsky
Jon Lewis | I route
Senior Network Engineer | therefore you are
Atlantic Net |
_________ http://www.lewis.org/~jlewis/pgp for PGP
public key_________

__________________________________
Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

Thanks for the info Rodney,

On Fri, Mar 25, 2005 at 11:49:41PM -0800, Lawrence

Post by Lawrence Wong
Which version will this new 12.2S image belong to?

The name hasn't been officially set yet.

Post by Lawrence Wong
On an offtopic issue, I was running 12.2(18)S on

my

Post by Lawrence Wong
7505 facing the Internet just fine but anything

from

Post by Lawrence Wong
12.2(18)S1 to S8 kills it outright with dbus

timeout

Post by Lawrence Wong
errors but 12.2(25)S to S3 works okay.

We are deferring 12.2(18)S8 for the 75xx.
Don't load it if you are running dCEF.
CSCdz84963
Sorry for the problems it caused.
Rodney

Post by Lawrence Wong

Post by Rodney Dunn
I heard there is a dCEF problem in that throttle
and they are working to defer it for the 75xx.
If you want MLPPP in 12.2S you had better wait a

bit

Post by Lawrence Wong

Post by Rodney Dunn
until the new 12.2S image comes out that

integrates

Post by Lawrence Wong

Post by Rodney Dunn
the MLPPP and MPLS stuff because a lot of MLPPP

bugs

Post by Lawrence Wong

Post by Rodney Dunn
are fixed there.
For now, go with 12.3 or 12.0S.
Rodney
On Thu, Mar 24, 2005 at 09:46:06AM -0800, Chris

Post by Chris Cappuccio

Post by Richard J. Sears
Would 7513's running 12.2(27) have problems

with

Post by Lawrence Wong

Post by Rodney Dunn
MLPPP...? We had to

Post by Chris Cappuccio

Post by Richard J. Sears
convert all of our MLPPP customers to ip

load

Post by Lawrence Wong

Post by Rodney Dunn
sharing per packet because

Post by Chris Cappuccio

Post by Richard J. Sears
we would lose the mlppp links and the only

way

Post by Lawrence Wong

Post by Rodney Dunn
to get them back was to

Post by Chris Cappuccio

Post by Richard J. Sears
reboot the router. I would like to use

MLPPP,

Post by Lawrence Wong

Post by Rodney Dunn
but I (like many) have had

Post by Chris Cappuccio
Shit, I tried 12.2(18)S8 and it broke all my

MLPPP

Post by Lawrence Wong

Post by Rodney Dunn
connections that worked

Post by Chris Cappuccio
fine on 12.2(18)S4. The situation is

7507/RSP4....everything was up/up upon

Post by Chris Cappuccio
moving to S8, but no traffic passed, and ppp

debug

Post by Lawrence Wong

Post by Rodney Dunn
showed continuous "PROTREJ"

Post by Chris Cappuccio
messages, as in, something is really broken.

TAC

Post by Lawrence Wong

Post by Rodney Dunn
is being slow to respond :)

Post by Chris Cappuccio
--
"All creativity is an extended form of a joke.

Most creativity is a transition

Post by Chris Cappuccio
from one context into another where things are

more surprising." - Alan Kay

_______________________________________________

Post by Lawrence Wong

Post by Rodney Dunn

Post by Chris Cappuccio
cisco-nsp mailing list

https://puck.nether.net/mailman/listinfo/cisco-nsp

Post by Lawrence Wong

Post by Rodney Dunn

Post by Chris Cappuccio
archive at

http://puck.nether.net/pipermail/cisco-nsp/
_______________________________________________
cisco-nsp mailing list

https://puck.nether.net/mailman/listinfo/cisco-nsp

Post by Lawrence Wong

Post by Rodney Dunn
archive at
http://puck.nether.net/pipermail/cisco-nsp/

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam

protection around

Post by Lawrence Wong
http://mail.yahoo.com

__________________________________
Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

How about 3640 running plain IOS and just running
MLPPP with no NAT, no IPFW, no nothing? I noticed eBay
selling quite a number of 3640 "cheap".

Post by Lawrence Wong
May I know if that is a 3620 or 3640? Any issues

in scaling up to 4 x

Post by Lawrence Wong
T1/E1 ?

It is a 3620.
I am pretty sure it could scale just fine if you
were not doing NAT on it
(mine is). But as for the 2500 platform that
definitely will not do 2
t1's pegged.
It will, however, do 2 T1's if the bandwidth is
pretty asymmectrical
(mostly
incoming or mostly outgoing)
Ted

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Ted Mittelstaedt]

Post by Lawrence Wong
How about 3640 running plain IOS and just running
MLPPP with no NAT, no IPFW, no nothing? I noticed eBay
selling quite a number of 3640 "cheap".

Post by Ted Mittelstaedt

May I know if that is a 3620 or 3640? Any issues in scaling up to 4
x T1/E1 ?

It is a 3620.
I am pretty sure it could scale just fine if you
were not doing NAT on it
(mine is). But as for the 2500 platform that
definitely will not do 2
t1's pegged.
It will, however, do 2 T1's if the bandwidth is
pretty asymmectrical
(mostly
incoming or mostly outgoing)

Even better, the 3640 has a 100Mhz R4700 processor, the 3620 has an 80Mhz
R4700 processor.

Make sure to include the cost of Cisco Relicensing for Used Equipment
for the IP feature set your planning on running, and the cost of a Cisco
service contract so you have permission to run the most current IOS
version on it. And, remember that Cisco service on used equipment does
not
carry hardware support, so if it dies, you have to buy another one off
Ebay.

Ted

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Gert Doering]

Hi,

Post by Lawrence Wong
How about 3640 running plain IOS and just running
MLPPP with no NAT, no IPFW, no nothing? I noticed eBay
selling quite a number of 3640 "cheap".

The 3640 is somewhat faster than the 3620 (100 MHz vs. 80 MHz R4700,
225 R5271 in the 3660), and can take up to 128 Mb of RAM.

The disadvantage is that it comes without an onboard ethernet interface,
and then NM-1FE-TX seems to run at about the price of a used 2621, at
least over here in Germany.

gert

[Ted Mittelstaedt]

Post by Gert Doering
Hi,

Post by Lawrence Wong
How about 3640 running plain IOS and just running
MLPPP with no NAT, no IPFW, no nothing? I noticed eBay
selling quite a number of 3640 "cheap".

The 3640 is somewhat faster than the 3620 (100 MHz vs. 80 MHz R4700,
225 R5271 in the 3660), and can take up to 128 Mb of RAM.
The disadvantage is that it comes without an onboard ethernet
interface, and then NM-1FE-TX seems to run at about the price of a
used 2621, at least over here in Germany.

Gert, he is probably looking at buying something like this:

http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&category=51203&item=5762764
500&rd=1

for $1000 USD, it already comes with FE and serial ports in it, all he
would need are cabling and external DSUs. And it would be faster than a
2621 non-XM.

Ted

_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Gert Doering]

Hi,

Post by Ted Mittelstaedt
http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&category=51203&item=5762764
500&rd=1
for $1000 USD, it already comes with FE and serial ports in it, all he
would need are cabling and external DSUs.

Indeed, that's an interesting price. OTOH, the fact that there are 0 bids
yet speaks for itself, doesn't it?.

Post by Ted Mittelstaedt
And it would be faster than a 2621 non-XM.

OTOH used 2621XMs with warranty can be had for about 700 EUR (over here
in Germany), and the WIC-1T goes for about 50 EUR - so $1000 for a 3640 is
not *such* a good price...

Then, again, memory for 3640 is cheaper than for 2621... so it all depends
on local preference.

I'm not opposing 3640s, though - I just seriously dislike 3620s. We own
three, and in retrospective, this was a not a smart buy.

gert

[Lawrence Wong]

Post by Lawrence Wong

Post by Lawrence Wong
Any idea how many E1 MLPPP is supported on the

2621? I
The 2621 is quasi-identical to the 2620. I have
first hand experience
with trying to upgrade a 2xE1 MP bundle on such box
to 3xE1. It did
work in principle, but the CPU load on this box
reached peaks beyond
60% when that bundle got fully loaded with traffic.
This caused an
unusual amount of RTT jitter and response loss on
pings to that router
what finally convinced us to give up the MP and
retry with CEF load
sharing per-packet. The box has 15 to 20% less CPU
load this way.
However, keep in mind that this was with a late
12.1, not really a
recent IOS. As already pointed out by others here,
newer IOSs can do
MP CEF switched, so the numbers might be better
there. This is something
I'd actually want to lab test soon as I'd really
like to get back to MP
there if possible.

Do let us know how things go if you get to test out in
your lab. It sure will be fun to see how much "soft"
engineering advanced over the versions.

Post by Lawrence Wong

Post by Lawrence Wong
don't think I will grow beyond 2 channels but if I

Two channels is working great. You get all the
advantage of MP including
perfect load balancing, guaranteed no reordering and
load balancing of
IP multicast (something I can't stop repeating you
*don't* ever get with
CEF per-packet).

That's one thing I like about MLPPP over others like
per packet load sharing, etc.

Post by Lawrence Wong

Post by Lawrence Wong
need to replace the 2514, it would be on the safe

side

Post by Lawrence Wong
to cater for 3-4 E1 MLPPP just in case (would a

3620

Post by Lawrence Wong
be good enough?) .

Forget 3620. Rather consider the newer 26xx models

I was thinking if the 3640s on used market will do the
trick? Looks like more powerful than the 3620.

Post by Lawrence Wong
or even 18xx/28xx. They
aren't even really expensive compared to former
gear. If ever going 3k
series, rather consider 37xx or 38xx, with exception
for the 3660.

The 37xx & 38xx looks very overkill for my "warehouse
to headoffice" connectivity. I was thinking of the
1841 but the Cisco brochure at
http://www.cisco.com/application/pdf/en/us/guest/products/ps5853/c1244/cdccont_0900aecd801738fa.pdf
says it's more capable of maxing out a 2Mbps link but
no mention of 2 x 2 or 3 x 2. Anyone played with these
new kids on the block yet?

Post by Lawrence Wong

Post by Lawrence Wong
I am thinking of using the VWIC-2MFT-G703 module

on

Post by Lawrence Wong
the 2621 to connect the individual links instead

of

Post by Lawrence Wong
getting external v.35 convertors.

Do you *really* need unframed E1? It's the only
additional virtue of the
-G703 over the -E1 and surely not worth the
significant price step except
you must have it. You can even stamp a G.704 frame
onto a line that is
provided unframed, so the only reason you must have
unframed is when
talking to a remote CSU/DSU that is limited to
unframed either by design
or by not beeing under your control.

I guess I need the G703 cos that's what the telco
drops in my lap.

Post by Lawrence Wong

Post by Lawrence Wong
Anyone has experience with it?

So far only with the -E1. Work like a charm. We've
got several 26xx
(including 2620) packed with those VWICs and they
are doing great.
For proper MP operation with 2xE1 and CBAC we prefer
the 2650 class,
but otherwise a 2620 can well deal with 4 individual
E1 interfaces at
full load without a tendency to fall over.

When running your bundles, are you running other
features like CBAC, NAT, etc that may increase CPU
overheads?


Thanks,

Post by Lawrence Wong
HTH,
Andre.
--
The _S_anta _C_laus _O_peration
or "how to turn a complete illusion into a
neverending money source"
-> Andre Beck +++ ABP-RIPE +++ IBH Prof. Dr.
Horn GmbH, Dresden <-
_______________________________________________
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at
http://puck.nether.net/pipermail/cisco-nsp/

__________________________________
Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Gert Doering]

Hi,

Post by Lawrence Wong
I guess I need the G703 cos that's what the telco
drops in my lap.

Every E1 is G.703.

The question is "will you need raw bits on G.703" (giving you 2048 kbit/s.
net data throughput) or "will you need 32x64kbit/s G.704 framing on top
of G.703, of which you can use 31x64 for data (1984 kbit/s. net throughput)"?

You can run G.703/G.704 framing on top of every E1 you get, except, as
Andre said before, if you need to talk to a CSU/DSU on the other end
that insists on running 2048 kbit/s. unframed G.703.

gert

[Christopher E. Brown]

Post by Lawrence Wong

Post by Andre Beck
Two channels is working great. You get all the
advantage of MP including
perfect load balancing, guaranteed no reordering and
load balancing of
IP multicast (something I can't stop repeating you
*don't* ever get with
CEF per-packet).

That's one thing I like about MLPPP over others like
per packet load sharing, etc.

Ya, but I would be alot happier if MLPPP was available at the DS3 level
on the GSR...

Ya I know, screwy need, but getting OC3s to Seattle on diverse fiber can
be a bit problematic (least around here).
--
------------------------------------------------------------------------
Christopher E. Brown <***@acsalaska.net> desk [phone]
cell [phone]
IP Engineer - ACS
------------------------------------------------------------------------
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

Hi,
On Mon, Mar 28, 2005 at 09:25:36AM -0800, Lawrence

Post by Lawrence Wong
How about 3640 running plain IOS and just running
MLPPP with no NAT, no IPFW, no nothing? I noticed

eBay

Post by Lawrence Wong
selling quite a number of 3640 "cheap".

The 3640 is somewhat faster than the 3620 (100 MHz
vs. 80 MHz R4700,
225 R5271 in the 3660), and can take up to 128 Mb of
RAM.
The disadvantage is that it comes without an onboard
ethernet interface,
and then NM-1FE-TX seems to run at about the price
of a used 2621, at
least over here in Germany.

Hmm... any idea what's the processor like on the 2621
and 2621XM? If I recall correctly, it is about 66Mhz
range or slower, which means a 3620/40 is actually
more powerful (except that they don't come with
onboard FEs)? Or is there some other factor that
affects the performance of a 26xx vs 36xx ?

Best regards,




__________________________________
Do you Yahoo!?
Take Yahoo! Mail with you! Get it on your mobile phone.
http://mobile.yahoo.com/maildemo
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

Hi,
On Mon, Mar 28, 2005 at 09:33:33AM -0800, Lawrence

Post by Lawrence Wong
I guess I need the G703 cos that's what the telco
drops in my lap.

Every E1 is G.703.
The question is "will you need raw bits on G.703"
(giving you 2048 kbit/s.
net data throughput) or "will you need 32x64kbit/s
G.704 framing on top
of G.703, of which you can use 31x64 for data (1984
kbit/s. net throughput)"?
You can run G.703/G.704 framing on top of every E1
you get, except, as
Andre said before, if you need to talk to a CSU/DSU
on the other end
that insists on running 2048 kbit/s. unframed G.703.

Thanks for the details. Is it possible for me to get a
pair of VWIC-1MFT-G703 (one each on two different
2621) to talk to each other so that I can do some
testing?

I know it is possible to do a back to back cross over
between "normal" serial interfaces like that found on
the 2514 and WIC-2T. But what about the
VWIC-1MFT-G703? Anyone tried before using RJ45 cables
to connect routers back to back? Any special settings
to look out for?

Thanks,



__________________________________
Do you Yahoo!?
Yahoo! Mail - Easier than ever with enhanced search. Learn more.
http://info.mail.yahoo.com/mail_250
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Gert Doering]

Hi,

Post by Lawrence Wong
Thanks for the details. Is it possible for me to get a
pair of VWIC-1MFT-G703 (one each on two different
2621) to talk to each other so that I can do some
testing?

Yes. Crossover wiring, 1+2 <-> 4+5, set one side to internal clocking,
and off you go.

gert

[Lawrence Wong]

Hi,
On Mon, Mar 28, 2005 at 09:33:33AM -0800, Lawrence

Post by Lawrence Wong
I guess I need the G703 cos that's what the telco
drops in my lap.

Every E1 is G.703.
The question is "will you need raw bits on G.703"
(giving you 2048 kbit/s.
net data throughput) or "will you need 32x64kbit/s
G.704 framing on top
of G.703, of which you can use 31x64 for data (1984
kbit/s. net throughput)"?
You can run G.703/G.704 framing on top of every E1
you get, except, as
Andre said before, if you need to talk to a CSU/DSU
on the other end
that insists on running 2048 kbit/s. unframed G.703.

Thanks for the details. Is it possible for me to get a
pair of VWIC-1MFT-G703 (one each on two different
2621) to talk to each other so that I can do some
testing?

I know it is possible to do a back to back cross over
between "normal" serial interfaces like that found on
the 2514 and WIC-2T. But what about the
VWIC-1MFT-G703? Anyone tried before using RJ45 cables
to connect routers back to back? Any special settings
to look out for?

Thanks,




__________________________________
Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

[Lawrence Wong]

Thanks for the clarification!

Post by Yury Yaroshevsky

Post by Lawrence Wong
Am I correct to say that if using per packet load
sharing, the maximum throughput at any one time

will

Post by Lawrence Wong
be the max of the individual link and not the

bundle

Post by Lawrence Wong
as a whole?

You have it inverted.
per-packet shortcuts CEF (does a lookup for each
packet, spreads across
pipes)
per-flow (the default) uses CEF normally (one flow,
one pipe)
--

------------------------------------------------------------------------

Post by Yury Yaroshevsky
desk [phone]
cell [phone]
IP Engineer - ACS

------------------------------------------------------------------------
__________________________________
Do you Yahoo!?
Take Yahoo! Mail with you! Get it on your mobile phone.
http://mobile.yahoo.com/maildemo
_______________________________________________
cisco-nsp mailing list cisco-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/