[67514] in North American Network Operators' Group
Re: 7500 and DCEF
daemon@ATHENA.MIT.EDU (Hank Nussbacher)
Wed Feb 11 00:28:17 2004
Date: Wed, 11 Feb 2004 07:26:39 +0200
To: nanog@merit.edu
From: Hank Nussbacher <hank@att.net.il>
Errors-To: owner-nanog-outgoing@merit.edu
On Wed, 07 Jan 2004, I posted:
>> They recently refreshed the platform with RSP16, VIP8, and MX. It's still
>> a viable platform for many medium size providers.
>
>As an exercise see if you can determine when this 7513:
>http://noc.ilan.net.il/stats/ILAN-CPU/new-gp-cpu.html
>swapped from an RSP8 to an RSP16 in the past 2 months.
After posting this a month ago, I felt I also needed to post a follow-up
on the matter.
If one looks at the CPU graph now, you will see that I was able to get
the RSP16 to drop from 54% to 43% via altering all out ACLs from named
to numbered. This is due to a bugid: CSCec07566 (Packets for interface
with named ACL are not dcef switched) which affects many 12.2S versions.
Next I added "ip multicast-routing distributed" and "ip mroute-cache
distributed" to all multicast enabled interfaces, A reload 6 days
ago caused the CPU to drop from 43% to 22% - probably due to some
unknown reason - whereby all the DCEF changes were not taking hold.
Lastly, on all slow serial interfaces (2Mb/sec and under), I changed the
queuing to "fair-queue" and that caused DCEF to kick in and bought
another 10% reduction. I am now at around 10% CPU and happy.
I'd like to thank Rodney Dunn of Cisco for suggesting many DCEF fixes
and configuration optimizations.
-Hank
