[c-nsp] IOS image for 3640 w/ DS3 module

[Mikael Abrahamsson]

On Fri, 8 Sep 2006, Kevin Graham wrote:

> What's CPU utilization look like? You're going to have to run very
> basic interface configs and hope for large packets to get a 3640 to
> handle a full DS3.

Well, I actually have to agree with Mark, the original poster. I have 
consistantly verified performance differences between different IOSes on 
the 3640, though this was with HSSI/Kentrox to go over DS3. It was at 
least a factor 2 difference, whereas without features with just a few FTP 
sessions the 3640 would go at full CPU on some versions, but not with 
others. This was just basic setup with WFQ on the interfaces and basic IP 
routing.

The strange thing was that there was a difference between a 12.1 config 
booted on 12.0S, and a 12.0S config booted on 12.0S. So basically, if I 
booted 12.1 (or 12.2), did a "write", and rebooted into 12.0(28)S, it 
worked properly. If I wrote the config on the 12.0 and rebooted, I got 
worse performance.

The solution in question is no longer active so I can't log a TAC case on 
it, but it was quite strange, but 100% reproducable, I did it 5 times 
in a row to really verify the problem.

So basically what I'm saying, is that the 3640 hardware can fill a DS3 
using HSSI and certain IOS version, where traffic is mostly a few TCP 
sessions (so mostly large packets and ACKs, where the large packets are 
going one way and the ACKs are going the other), and there seems to be 
some IOS weirdness going on that makes this differ between versions.

-- 
Mikael Abrahamsson    email: swmike at swm.pp.se