Phil,
The reason for the lengthy exposition was that most people who aren't actively
involved in QoS work usually don't understand the issues. My view was (and
still is) that a "level set" is needed. Unfortunately, there is not one RFC
(and I haven't found a good single paper in the literature) that one can point
to that has information of this nature.
QoS stuff is somewhat abstract, because unless one starts at first principle,
one is likely to get it wrong. I disagree that "this has kept it from making
progress in the real world." There are other reasons, which rate a
side-discussion in San Diego but are off topic for this list.
I don't think that the amount of exposiontion in my proposed text is
inconsistent with the amount of exposition on, say TCP. There probably is
some material which could be cut back, and if you strongly need to do that, it
could be done. However, I really think your current version threw out the
baby with the bathwater. It also retained some material that is at least
misleading if not incorrect, namely the first couple of paragraphs.
WRT subnet designers, in addition to the tutorial material, I think we need at
a minimum to make the following points:
-- QoS is quantitative and measureable. QoS != priority.
-- Don't waste your time on QoS classes; use network mechanisms like RSVP to
negotiate
-- Don't assume that just setting a few bits in the IP header (TOS or
Diffserv) solve your QoS problems
-- QoS guarantees require cooperating mechanisms that play together as a
system. Some of these are not stateless.
I'm not able to do much on this for the next few days. We can talk about this
in San Diego, or if you can be specific as to how much material you think
needs to be cut out, I can try to rework.
Dan
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 09:12:25 EST