[c-nsp] Your opinions on router throughput

[Gert Doering]

Hi,

On Sat, Oct 28, 2006 at 03:22:33PM -0500, Tony Varriale wrote:
> There is nothing wrong with the device in question....let alone seriously.  
> It is asked to do a lot of stuff.

If it's moving "10 Mbit/s", it's not doing "lots of stuff".

10 Mbit/s. of imix traffic is something a 3640 can handle without breaking 
sweat.

> Is there a particular reason you call "something is wrong" or
> "open a TAC case" every time a performance measure falls outside
> your expected levels?  

Yes - because it's not helpful if people give advice on this list based
on misconfigurations, and experience shows that almost every case of
a Cisco device experiencing extremely unusually high load / bad
performance is either cause due to software bugs, or misconfiguration
(like "turning off CEF because someone said CEF sucks").

Note that I never suggested "open a TAC case" in this particular thread,
and going through all my postings I've ever sent to this list, I'm fairly
sure you can count the total postings where I've *ever* suggested that
on two hands...

> Especially when you are not familiar with
> the configuration or duties the router performs...

The poster that started the thread had very specific questions about
a very specific environment.

So if someone is posting "a NPE300 will be at 70% with 10 Mbit/s. of
load" (without specifying that there are extraordinary circumstances)
it's certainly not matching expected performance - and unless this person
is actually specifying the difference between his environment and the
enviroment that the original poster is interested in, the whole comment
is fairly useless.

Of course I can get a NPE300 to 70% load with 10 Mbit/s of traffic - turn
off route caching and CEF, and turn on ip accounting, ACLs, ACL logging,
and netflow.  Or so.

> Can you provide actual performance of an NPE300 in a production 
> environment with x,y and z turned on?  I can...and that's what I offered.

Actually, you didn't, because you did not tell what "x, y and z" you've
turned on.


To come back to useful figures - I've just checked one of our NPE300s running
as LNS, doing L2TP termination, running 12.3(9a), IPv4 and IPv6, netflow
accounting on all vaccess interfaces.  It peaks at about 20-25 Mbit/s. of 
traffic, with about 40%-45% CPU peak.

Given the fact that a not-so-small percentage of the L2TP traffic is
fragmented (due to not all traffic being TCP and thus being mss-adjusted)
this can be considered "normal" - but it's higher than the CPU load
would be for a box that just shoves packets, without tunneling of any
sorts, and without netflow.

gert

-- 
USENET is *not* the non-clickable part of WWW!
                                                           //www.muc.de/~gert/
Gert Doering - Munich, Germany                             gert at greenie.muc.de
fax: +49-89-35655025                        gert at net.informatik.tu-muenchen.de