[j-nsp] TCAM full on EX8200?

Pavel Lunin plunin at senetsy.ru
Thu Oct 20 10:32:02 EDT 2011


>> This is where the number of labels comes into play. If we
>> talk about LSR for not that huge IPS (having not that
>> much of core LSPs), I'm afraid, this can require to get
>> back to the old good conception of FEC per prefix :)
> When we were small and using Cisco 7200's as BGP-free core
> routers, we were load balancing just fine. And this was a
> network doing less than 500Mbps of aggregated traffic.
>
> Of course, things are far more different nowadays :-).

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.

When it's not a problem for a software-based platform, I'm afraid, cheap 
network processors, used in switches, can't do that. Thus I'm not sure, 
they are sufficient for a good LSR. Despite they have plenty of power 
for lookups.


More information about the juniper-nsp mailing list