On Sep 1, 2005, at 6:43 AM, Dave Lag wrote:

>> There is a lot of interest in latency/QoS for packet networks (again
>> driving up prices), but it's not clear to me to what extent that is
>> truly important

> Well, voice and latency/unpredictability don't mix.

not naively, anyway.  I sure wish more people were working on 
intelligent
packetization of speech, for instance, so that packet boundries would
occur at the natural pauses in speech.  Like I complained before, too
much effort is being spent on duplicating the characteristics of the
existing infrastructure, rather than exploring the possibilities of
the new technology.  fer instance, Katrina hits and all the VoIP
infrastructure switches codecs and reduces bandwidth per call at
noticeable degradation in voice quality, but noticeable increase
in the call volume that can be handled...

> Either you profile all your customers and size accordingly or traffic 
> shape at the endpoints or keep separate networks at the transport 
> level or ?

Well, SOME of all the above, probably.  Currently, people seem to think
that they must do ALL of those (plus QoS) simultaneously...

>
> Bandwidth isn't free, otherwise our ISPs wouldn't be throttling
> all the P2P packets :)

A certain amount of bandwidth is nearly free.  Upgrading the devices
on the ends of your fiber is very inexpensive compared to installing
the fiber in the first place.  And the same CAT5 that I installed for
10baseT seems to be working pretty good at 1000Mbps these days, with
upgrades that were at worst "rather inexpensive" at the ends.

But the question isn't whether bandwidth is free; the question is
whether added bandwidth is cheaper than deploying services that make
the existing bandwidth more useful for low-latency protocols...

BillW

-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist