[Asterisk-Users] QoS anyone?

[Rich Adamson]

> Rich,
> I would be surprised to find this.  Typically ISP's will reset all QOS 
> settings to 0 either on your CPE router if they manage it or on the 
> aggregation router your circuit is connected to.  Almost always if they 
> support DSCP/TOS matching and priority queuing in the core of their 
> network it's part of an extra charge service. If they don't do any 
> priority queuing then they typical will just leave it alone and ignore it.

Having spent a number of years doing professional network performance and
security work (and extremely heavy into protocol analysis, etc), I was _very_
surprised as well. Not suggesting all backbone providers support it.

We spent a fair number of billable hours attempting to diagnose a dsl vs
DS3 problem recently. The issue boiled down to: voip worked fine over a
relatively low bandwidth dsl circuit, but failed (poor quality) over a
dedicated DS3 (via a certain 'major' provider). In case anyone missed that,
it was a 'single' PC on the DS3 with unacceptable voip to another D3 site
about 800 miles away! No other traffic.  After analyzing the packet trace 
data, we found the tos bits were being honored on the dsl but not on the 
ds3. Does that suggest all backbone ISPs support it? Absolutely not.

(This happened to be a major trade show where the end result had to be
of recognizable quality.)

But, I can absolutely assure you we know _exactly_ what we're doing in 
analyzing the trace files. What we obviously don't know is all the 
components involved within the cloud. Trace routes were used to confirm
the paths, etc.

The end result was the DS3 provider was truly handling the tos bits, while
the dsl provider was treating tos as non-existent. (In this particular case,
one of the end nodes was an XP box, and someone had (unknowningly) unchecked
the QoS setting. Repeated tests that involved nothing more then flipping
the option verified the end result over and over. Extremely surprised!

As a non-believer, I played around early this morning setting asterisk tos
bits "to" a C7960 roughly 12 hops across the Internet. The testing was not
scientific by any means. However, there was a very noticable difference
between tos bits on verses off. Could there have been something else impacting
this particular test? Sure, but will be replicating the test using a packet
sniffer (again) to validate the results under different conditions.

Surpised? Yes; believer? not totally, yet; ISP/path dependent? probably.

Rich