Re: [j-nsp] BGP memory consumption

From: john heasley (heas@shrubbery.net)
Date: Fri Feb 16 2001 - 13:45:18 EST


Fri, Feb 16, 2001 at 08:44:07AM -0600, Tony Mumm:
> Our peer there is also doing similar memory usage..roughly 117 MB for
> that many peers. The question I counter with, is PFE memory
> requirements. It appears that the pfe uses 8 MB of super high
> quality RAM to record next-hops for the IP2 ASIC (assumption on my
> part). From show pfe route summary, it appears a router is using
> 5.8 of this to run ~100k routes. Now I don't know how this is going
> to scale in relation to the number of routes, but if you just use a
> simple one to one relationship the router will top out at around 140k
> routes. This number is based on roughly 59 bits per route in
> memory... again, entirely guesstimates.
>
> Obviously the rest of the pfe subsystem has 64 MB of ram...but I
> don't know how this applies..
>
> I'm not trying to be negative here, just want to see if anyone else
> has thought of this?

dont confuse pkt/fabric memory w/ fib memory.

> Thanks,
>
> Tony
>
> -----Original Message-----
> From: nicholas harteau [mailto:nrh@ikami.com]
> Sent: Thursday, February 15, 2001 11:49 PM
> To: Josh Richards
> Cc: juniper-nsp@puck.nether.net
> Subject: Re: [j-nsp] BGP memory consumption
>
>
>
> Your juniper will have RIBs for each peer/protocol pair. In my
> experience the memory use scales linearly all else being equal.
> Also,
> I'd be hard-pressed to find a scenario outside of a lab where you
> will
> use more than the 768M of ram that the junipers support these days.
> My
> AADS router is using 128M with 55 peers and mbgp running as well as
> bgp.
>
> Josh Richards wrote:
> > How does Juniper's BGP memory consumption compare to, say, a Cisco?
> >
> >
> > While the exact answer obviously varies depending on the peering
> > sessions, what magnitude of memory consumption can I expect to be
> > necessary for the first full transit view (~100,000 prefixes) and
> > each successive additional transit view of similar size?
> >
> > Is the memory consumption in such a way that I can assume that a
> > peer with, say, only 10,000 prefixes is going to consume 10% of
> > the amount of memory as the 100,000 prefix peer?
> >
> > Damn, I can't wait to get a few of these things in the lab...
> >
> > -jr
> >
> > ----
> > Josh Richards [JTR38/JR539-ARIN]
> > <jrichard@geekresearch.com/cubicle.net/fix.net/freedom.gen.ca.us>
> > Geek Research LLC - <URL:http://www.geekresearch.com/>
> > IP Network Engineering and Consulting
> >
>
> --
> nicholas harteau
> nrh@ikami.com
>



This archive was generated by hypermail 2b29 : Mon Aug 05 2002 - 10:42:40 EDT