Re: critique of NANOG meeting

[Avi Freedman <freedman () netaxs com>]

> Load balancing is a funny stuff.  Its utility is very limited
> by the fact that you've got to keep packets in sequence to
> avoid triggering TCP fast retransmits.
>
> I.e. with ciscos you can only do load-balancing on paths
> (i.e. load one circuit with traffic to ISP A and another to ISP B
> etc) or on destination networks (that has a potential of causing
> very skewed load distribution).

I admit I'm not familiar with the impact of 'ip route-caching' 
tens of thousands of equal-cost paths at dual-HSSI speed.  I was
told by some at Cisco that it wouldn't be a problem.

> Before i left Sprint i produced a plan for them to increase
> capacity at least four-fold using combination of load balancing
> and hot-potato intra-backbone routing.  Going to OC-3s is
> a lot simplier, though :)

True, it's simpler - but one does what one needs to do to stay
alive.  Or one should, at least :)

> --vadim

Avi
- - - - - - - - - - - - - - - - -