[e2e] Open the floodgate - back to 1st principles

[Jonathan Stone]

In message <83AB0942FD087D499DF2DD5CEE1B61330856D7 at cacexc06.americas.cpqcorp.net>"Ghanwani, Anoop" write

[...]

>You would need multiple TCP-level flows
>to see any benefit from link aggregation (at least the 
>way it's defined in IEEE 802.3).

Yes. If memory serves, the IEEE link-aggregation spec requires all
frames in a MAC-level flow (a unicast source-MAC/destination-MAC pair)
to be forwarded over the same link in an aggregation group.  The spec
explains that this done is to avoid reordering of frames within a
`flow'. Clearly, somebody gave them the impression that reordering
traffic was a Very Bad Thing.

Various switch/router vendors do provide link aggregation options for
non-IEEE-conformant queuing (e.g., round-robin across all frames).
Round-robin queuing can and does give benefit to single TCP
flows. With the proviso that the single flow is coming in over a
fatter pipe than the individual components of the aggregated
link. Last I looked, I couldn't find a link-aggregation implementation
for end-hosts that will `round-robin' a single TCP flow across the
aggregated links.  (I'd appreciate hearing of any).