[c-nsp] bgp scalability C7600

[Phil Bedard]

We are connected to a Level3 7600 at one location, when we turned it up a few months ago it took almost 30 minutes to send us a full table.

Phil

-----Original Message-----
From: "Gert Doering" <gert at greenie.muc.de>
Sent: ‎2/‎6/‎2015 1:52 PM
To: "james list" <jameslist72 at gmail.com>
Cc: "cisco-nsp at puck.nether.net" <cisco-nsp at puck.nether.net>
Subject: Re: [c-nsp] bgp scalability C7600

Hi,

On Fri, Feb 06, 2015 at 03:16:26PM +0100, james list wrote:
> do anybody have numbers in terms of BGP sessions scalability oin C7600
> SUP-720 ?

"not that great"...

Ours at DE-CIX has a handful of iBGP sessions and about 150 eBGP sessions
to IXP participants, and if that interface flaps, it will hickup for 
about *1 hour* until everything is stable again.

Effectively it depends on

 - number of sessions
 - number of prefixes on each session ("10 each" or "50.000")
 - how complicated your inbound and outbout policy is
    (our policy is slightly too complicated, with as-path matches which
     are not exactly performance efficient)
 - whether peers can be grouped into update-groups (= same export policy)
 - keepalive timers your peers have configured
    (the main issue is "CPU busy -> keepalives not answered in time ->
     session bouncing -> more CPU busy", which is made worse by short 
     keepalive timers)

We're not deploying Sup720s for anything with "lots of BGP" anymore, and
the box in question will be replaced with an ASR9001 any day now, which
is just laughing its NPUs off on that BGP load... ("BGP convergence in 30
seconds.  done.  anything more interesting to do?  Any slow peer I could
nuke with outgoing updates sent over too fast for it?").

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