Hi all. I've been going over the PILC mail since the last IETF and am
currently revising the LINK document before Friday's deadline. The old
obsolete text is now gone from my copy; the question is how to replace it.
In re-reading the discussion started by Dan Grossman's suggested new
text for the QoS section, a few things occurred to me.
I have always been struck by how discussions about QoS inevitably
become extremely abstract, and I believe this has kept QoS from making
much concrete progress in the real world.
For example, Dan's section begins with a fairly lengthy (and abstract)
review of current differentiated services work and ends with a single
paragraph that dealt directly with link design issues. Given that not
much of a consensus was reached in the ensuing list discussion, I'm
inclined to leave out much of the review stuff in favor of pointers to
existing documents, and to keep his last section.
Since this is a document intended for subnetwork designers, any
recommendations we make about QoS should be in concrete terms that
subnetwork designers already understand. I still believe we should
simply point out that many subnetworks have "knobs" that can make
qualitative tradeoffs between competing measures of performance, that
applications differ in which performance measures they consider
important, give a few representative examples of such knobs, and
recommend that the designers not preclude the ability of some future
IP QoS mechanism to control them should the QoS folks ever figure out
what they really want to do.
Any comments?
Phil
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 09:12:25 EST