[c-nsp] Cisco memory requirements - Best practices

[jp]

I can do two full tables very reliably with 80MB free on an RSP4/256M 
using a recent 12.2 series IOS. I don't do anything else on it though. 
Ethernet and T3 upstreams, ethernet to the LAN.

We use other boxes to do everything else, choosing whatever is best 
suited for the task.

On Thu, Mar 16, 2006 at 11:47:18AM -0500, David Coulson wrote:
> I'm trying to get together some metrics following the issues yesterday
> with our 7500s. Right now we're just pulling <=/22 from two peers (about
> 66k routes) and we're running about 86Mb free of 256Mb.
> 
> >From a practical perspective, what is a reasonable amount of memory to
> have available for things such as BGP soft resets, HSRP additions, and
> so forth? We run both our 7507/RSP4 with dot1q, HSRP, BVIs and spanning
> tree enabled on the bridge-groups. From what info I can find on the
> router, these features don't seem to eat up much memory, however as we
> made a significant number of config changes in the last 15 days (adding
> bridges/BVIs, adding about 10 HSRP groups), the issue of memory
> fragmentation is a major concern to me.
> 
> I guess options are:
> 
> 1) Reduce size of BGP feeds to fit memory - Easy, and what we're doing
> right now.
> 2) Offload some process-based features (HSRP, etc) onto 7206s and just
> leave the 7500s as border routes doing BGP. This doesn't really solve
> the problem that we potentially can't cram two full tables into the
> RSP4s at some point, even with nothing else running. Plus I don't really
> think these features really consume a whole lot of memory -
> Fragmentation from changes is my concern here.
> 3) Use 7206s as border routers and use the 7500s for core - Seems messy,
> plus we'd have to run NPE-G1s, which brings me to #4
> 4) Replace the RSP4s with RSP16s (RSP8 maxes at 256Mb, right?)
> 5) Pull a single full table into each of our 7507s and do limited iBGP
> routes between the two. So, we'd pull 180k routes from each transit
> peer, then share 66k (<=/22) between the two routers. This would solve
> the problem with some of our downstreams who want full tables, but may
> reduce the overall memory usage on the routers by avoiding two full 180k
> tables. Backup the whole thing with default routes to transit peers and
> a lower preference route to the other router.
> 
> Another question related to memory - How much RSP memory am I losing by
> having various VIPs and PAs in the routers? If I pull anything that is
> not active, is that going to save me anything worthwhile? If I get rid
> of junk cards like FEIP, is it going to improve matters?
> 
> David
> 
> _______________________________________________
> 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/

-- 
/*
Jason Philbrook   |   Midcoast Internet Solutions - Internet Access,
    KB1IOJ        |  Hosting, and TCP-IP Networks for Midcoast Maine
 http://f64.nu/   |             http://www.midcoast.com/
*/