Re: Handling Packet Misordering?

[Fred Baker <[e-mail]>]

At 01:11 PM 9/9/2002 -0700, Rick Jones wrote:
>Wouldn't four or more segments outstanding on a
>session/connection/flow/whatever :) being relatively unusual imply that
>there would relatively little improvment (for that
>session/connection/flow/whatever) from a load balancing scheme that
>spreads packets of a given session/connection/flow/whatever across
>multiple links?

it does suggest that.

>Should we distinguish between eliminating reordering and not doing
>things that may reorder as a frequent matter of course?

I'm not sure that we encourage reordering "as a frequent matter of course" 
now. Many routers have the option of distributing traffic round-robin on a 
multi-path route, but most that I am aware of default to sending any 
specific microflow along a specific sub-path of that route, by such 
techniques as hashing the flow signature (source/destination address and 
perhaps more) and using the hash to deterministically select a route. The 
other behaviors that can cause reordering, such as changing routes, happen 
infrequently and are presumably necessary to implement a routing policy or 
avoid something worse, such as not having a route that works.

What I understood the original request to be was "eliminate reordering". 
That is what I have been addressing.