Re: Alternative to BGP-4 for multihoming?

[Vern Paxson <vern () ee lbl gov>]

> TCP performance is affected by congestion symmetry, since TCP uses the
> spacing of ACK packets to control the spacing of data packets.

True, but this is a second order effect.  The strongest effects are
(1) loss rate on the forward path, and (2) bottleneck throughput on the
forward path (these two being related).  TCP is much more tolerant of lost
acks than lost packets - you have to lose an entire flight of acks before
it really kills you.

So I suspect the match-the-outgoing-hop-with-the-incoming-hop approach
is working better for some other reason.  Perhaps it's reflecting that
Web surfing is in fact bidirectional, and often not all that assymetrically
(a small request fetches a small item).  In that case, congestion symmetry
will help, since both directions are forward paths at some point.

                Vern