RE: [nsp] CEF, BGP tables and memory

From: Martin, Christian (cmartin@gnilink.net)
Date: Mon Feb 12 2001 - 10:42:55 EST


We have problems running dCEF more often than CEF alone. Lots of old VIPs
with 32 MB RAM and a large Adj/FIB causes problems. This happens most on
routers with > 500 IDBs in use and/or a large ARP cache (> 5000 entries). I
agree that 128MB will not last for long. Also, the 7206 NPE uses
particle-based memory, which appears slightly more efficient than the RSP
system in terms of fragmentation management. Perform a show proc mem and
look for the amount of memory being held by *dead* processes. If this
number is over 10MB, you should reboot.

regards,
chris

 

>
> A 7206/NPE150 I have is managing with 128M RAM, 4 full feeds,
> and no CEF,
> with 14 Mbytes free (largest chunk 6Mbytes). CEF table is
> about 20Mbytes if
> you have a full routing table view, so you could carry 3 full
> feeds with a
> tight squeeze (but more realistic to carry only 2 feeds with
> CEF). Add
> soft-reconfig with lots of inbound policy, and 2 starts
> getting unrealistic.
>
> And that's with 12.0(13)S2 and the k4p version of the image.
>
> HTH,
>
> philip
> --
>



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:28 EDT