At 10:30 AM 12/18/2001, Sean Doran wrote:
>However, my feeling on pricing out edge QoS is that it should cost you
>money to have a provider do fancy-queueing for you...
>
> Likewise, my feeling on {diff,int}-serv is that core providers who have
> zero average length queues (i.e., ones with ample bandwidth) should
> simply be "QoS transparent",
>and simply never interpret the packet markings or participate in QoS
>negotiations
I'll modify those statements somewhat, but basically agree. Any service
that a provider does for you is something he should charge for; price is a
market issue. Any interface has a variety of strategies for providing QoS;
infinite bandwidth is one of them, engineered traffic is one that allows
him to accomplish the first with less actual investment is one of them,
drop management and queue management are among them when he can pick and
choose his applications (more of an enterprise issue). If the strategy you
have chosen is working for you, it's hard to argue that you picked the
wrong one.
Any interface that has bandwidth sufficient for mean+4*standard deviation
of its load during the busiest minute of the day (99.9% confidence of
sufficient bandwidth) is not contributing significantly to delay or jitter;
that's not religion, that's math. If those are the only links we're talking
about, I don't see much else to discuss. I understood us to be discussing
all of the service provider's links, not only his core links. I was wrong.
Is the subject of this RG "routing for infinite bandwidth links", or
"routing"? I could have sworn that I read recently some text about
considering all networks equivalent, and how having exactly two levels of
routing hierarchy and two classes of routing protocols was fundamentally
broken. I think Ran was getting at the notion that routing of applications
was interesting, not just routing of traffic. The reason that routing of
applications was interesting, if I understood him correctly, was that
applications run in places that don't have infinite bandwidth. Did I miss
something?
This archive was generated by hypermail 2b29 : Mon Aug 04 2003 - 04:10:03 EDT