[e2e] QoS vs Bandwidth Overprovisioning

[Vernon Schryver]

> From: Fred Baker <fred at cisco.com>

> >Has anyone (esp. major carriers) tried the simple, obvious approach of
> >honoring the old-style IP TOS bits?
>
> we went through quite a bit of effort on that a few years back, and finally 
> pulled them out of OSPF because the least monetary cost route, the highest 
> bandwidth route, the lowest delay route, and the most reliable route were 
> all the same route.

I don't understand what OSPF has to do with anything.  I'm not talking
about varying routes, but paying attention to the "low delay" IP TOS bits
while managing queues.  At higher (e.g. >=T3) speeds and so likely low
queueing delays, I don't think I mean re-ordering queue entries.  Instead,
prefer to drop packets that are not marked "low delay" when RED, queue
overflow, or whatever says that something must be dropped.

Honoring the "low delay" bits in queues for lower speed links by
re-ordering packets sounds like it could give radical improvments for
interactive (including voice) applications sharing links with bulk
traffic.  Over the years, I've found on low speed links it works great
to transmit all "low delay" packets before 90% of other packets (to
prevent their complete starvation).  On queue overflow I've dropped
the last packet in the queue after this re-ordering.

Again, the queue delays at OC-192 make talk of queue re-ordering
sound boring.  
At high speed, how can there be anyno latency or jitter problems unless
there is also packet loss?


Vernon Schryver    vjs at rhyolite.com