[c-nsp] BGP table growth and memory

[Gert Doering]

Hi,

On Mon, Jun 20, 2005 at 10:04:25AM -0700, Rick Ernst wrote:
> Is there any way of estimating base/operating memory requirements based on
> table size, number of peers, etc?  Is 256MB only OK "for another 2-3
> years", or am I turning myself into Chicken Little?

<rant>
256 Mb will last for "another 2-3 years" if people would finally start
to *care* what sort of garbage they announce into the routing table.

Unfortunately, these days it seems to be "pretty normal" to have a /19
and announce it as 32 x /24, and similar garbage that has no benefit
for anyone.

Studies by Geoff Houston and others show that the routing table could
be condensed a LOT without losing *any* information (that is: combining
only routing table entries where all BGP data [as path etc.] matches).
</rant>


As a rule of thumb, I'd give a 256MB/7200 box with 12.2(18)S that has 
full tables "but only iBGP peers" about another year.  A box with multiple 
full eBGP peers is likely to only last a few more months - especially if 
you use a more memory hungry IOS, like 12.2(25)S.  Dunno about 12.3 or 
recent 12.0S memory usage rates.  (And of course, this is pure speculation).

We've recently started upgrading all core routers that need to be 
touched anyway to 512 Mb or even 1 Gb in the NPE-G1 routers.  

Unfortunately, we still have a number of NPE-225 and NPE-300 routers, 
so we're also looking at alternatives - like "drop all routes more 
specific than a /20 from /8 blocks that live in other continents"
(what do I care about misguided traffic engineering ideas from some
of these networks?  They go over my upstream anyway).  

This is not unproblematic, though, so be aware of all consequences
(especially if you have BGP downstreams that expect to receive a *full*
BGP table from you).

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