[j-nsp] 802.3ad Question

[Richard A Steenbergen]

On Fri, Jan 15, 2010 at 09:18:01AM +0800, Mark Tinka wrote:
> Not quite - what I did say was:
> 
> 	"We're aggregating 3x Gig-E links between both devices, and 
> 	 it's working fairly well. Traffic distribution from the 
>        	 M7i to the 6500 is 1:1:1, while traffic
> 	 distribution from the 6500 to the M7i is 1:1:0.7 (but this
>          direction is governed by other Layer 3 devices
>          upstream of the 6500)."

The 6500/7600 platform (or at least the current generation of cards)
have only 8 hash buckets available, so if you use a non power-of-2
number of links like 3 above, what you're actually getting is 3:3:2. The
problem gets worse as the number of active links increases, so that if
you have 8 links at 1:1:1:1:1:1:1:1 and you lose one it becomes
1:1:1:1:1:1:2, effectively behaving like you only had 4 links instead of
7 (well, at least if you care about delivering half the packets on that
7th link at any rate).

I'm not aware of any major router vendor who still sells a product with
anything even beginning to approach this level of BAD. Juniper, Foundry,
Extreme, Force 10, etc, all have this pretty much nailed down.

-- 
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)