[nsp] NxT1 MLPPP

[David Curran]

You can bundle 8 links in a group on most platforms.  Some of the very
high-end distributed platforms will allow 12.  It should also be noted that
IOS will load balance, at most, across 8 paths (used to be 6).

Allow me to give some non-scientific anecdotes.  We had a lot Cisco 7206VXR
chassis with NPE400s running MLPPP links.  We ran OSPF and BGP across them
and some MPLS "stuff" over that.  We ended up having to convert all of them
to per-packet due to high CPU utilization and just general "weirdness".  The
boxes would run at 40% utilization even at "low tide".  And processor util.
would increase with load, which is not common with CEF.

After kicking this around for a while with cisco it became apparent that
some of the MPLS features we were running weren't 100% supported and that
was probably the cause.  There are some features suited for Multilink
interfaces and some that aren't.

The moral of the story is, as the previous poster said, don't skimp on
processor, don't expect to be able to run all protocols with any feature
over the bundle, and don't expect a 4xT1 group to actually run like a
factional DS3.  With those caveats in mind, MLPPP is a great technology.

If you're trying to decide which to use it may also be useful to consider
the traffic which will be passing over it.  If its going to be lots of small
packets, CEF with per-packet will run just as fast as MLPPP.  If its going
to be lots of large packets mixed in with other traffic you might benefit
from MLPPPs fragmentation and interleaving.  YMMV

Hope that helps,

-david

-----Original Message-----
From: cisco-nsp-bounces at puck.nether.net
[mailto:cisco-nsp-bounces at puck.nether.net]On Behalf Of Raymond, Steven
Sent: Wednesday, February 11, 2004 3:10 PM
To: Beprojects.com
Cc: Cisco Nsp
Subject: RE: [nsp] NxT1 MLPPP


Here is a quick bit of information I received from Cisco SE late 2002
pertaining to the 2620 (believe this is not XM performance info, which may
be better):


"You can either load balance via MLPPP or via per-packet CEF.  CPU usage
is a little lower with CEF, but not much.

4xT-1 on a 2620 at 64byte packets (worse case packet size) will get you
right at 6Mbps at 99% util.  With additional features such as routing
protocols and ACLS, performance decreases.

As you go up to the 2650 and 2690 series, the CPU gets faster, so your
performance is much better.

So, I suggest that if you want to guarantee the customer gets 6Mbps, go
with atleast the 2650.  This way you have a little room for routing
protocol and ACL processing.

Anymore feature requirements or T-1 density, you will have to jump to a
3700 series router."



-----Original Message-----
From: Beprojects.com [mailto:info at beprojects.com]
Sent: Wednesday, February 11, 2004 11:28 AM
To: Cisco Nsp
Subject: [nsp] NxT1 MLPPP


Does anybody have any performance numbers on NxT1 connectivity using MLPPP
on Cisco routers?  I am specifically looking for throughput on platforms
like the 2600 and 3700 (any flavors is fine).  How many T1's can you bundle
and how much throughput can it handle doing IP only with virtually no other
features turned on (cpu usage would be good as well).  Real world and/or lab
tests are both fine.   Thanks.

Peder

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