Re: [RPRWG] More comments on preemption
> Obviously, you're on the majority side of the preemption debate.
> But this time, I'm not trying to debate whether we should or should
> not do preemption, since everybody already form their own opioion
> about preemption in last round of debate. All I'm interested is to
> correct the misconception about preemption, so that everybody
> have a common base to make their own judgement.
>
> If you read requirement #5 and #6 carefully, there won't be a situation
> like "Multiple fragments from multiple packets sitting at a node,
> needing reassemblies". Only one assembly line per direction is
> needed. I just cannot emphasize enough to ask everybody not to
> take things out of context.
Right, most of us have worked on [ IP fragmentation / ATM Segmentation
and Reassembly / TCP segmentation ] and are reading in assumptions
from those earlier experiences. What is being discussed here is a
simple queue. An H packet can preempt exactly one packet at a time.
Only one buffer at the receiving node would be needed to rejoin the
pieces of that preempted packet.
Personally I don't believe preemption is desireable, but I'm glad
William is so ably pointing out the tradeoffs and issues to frame
the discussion.