[j-nsp] TCAM full on EX8200?

[Richard A Steenbergen]

On Thu, Oct 20, 2011 at 06:32:02PM +0400, Pavel Lunin wrote:
> 
> I meant that in order to do LB on labels alone (to have enough of 
> hash-keys for micro-flows), you need a large enough set of labels in 
> the core and more or less uniformly distributed traffic over these 
> labels. If you have, say, 10 PoPs and 90 core tunnels, it's very 
> probable that 20% of them carry 80% of traffic. But label-based hash 
> will share labels 50:50. This is why label alone is not sufficient for 
> limited set of LSPs and you need to construct hashes with more 
> parameters from payload.

Yes you need to look into the packet a little bit to hash well, but this 
isn't a difficult operation either (compared to holding a full table and 
doing longest prefix lookups at any rate). Honestly the BGP free core 
market probably isn't big enough to justify spinning a dedicated ASIC 
for an LSR, but you could probably get quite far with any of the 
existing commodity chips and a small amount of TCAM, if you had the 
right software support. I'm not a fan of the true BGP free core anyways, 
ICMP tunneling just confuses end users and results in nothing but 
support headaches. :)

FWIW EX8200 is actually kinda bad at multipath hashing too. It seems to 
use a very fixed hash seed which you can't manually alter (at least not 
that I've been able to figure out how to do), so it is EXTREMELY 
succeptible to multi-stage ECMP issues. We've seen things get as bad as 
30%/70% after two stages through two EX8200s of 2x ECMP each.

We haven't even bothered trying to do anything with MPLS and EX8200 
outside of the lab yet, as they're still missing a ton of important 
features that make it all pretty pointless right now (like say, ISIS TE 
extension support for example!).

-- 
Richard A Steenbergen <ras at e-gerbil.net>       http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)