RE: Real and Non-Real Time Traffic (was [WG-PIC:352] Re: FMM demo notes)

[Guy T Almes <>]

Jon,
 Two issues.
 First, if you did have several 5 Gb/s (attempted) TCP streams and if 
Reno-based algorithms were used by the TCP stacks, then you would get bad 
wild gyrations of queue size etc.  Very bad for voip, even apart from 
packet loss (which would also happen).

 Second, if you do TCP right, as with some of the newer approaches and/or 
implement RED appropriately at the bottleneck router (which is not trivial 
to do, by the way), then you might find that the elephants and voip would 
get along pretty well.

 Either way, though, given that performance bottlenecks are almost always 
at the edge, it might be hard for the Internet2 community to quickly induce 
chronic congestion in Abilene.  And if they did start to, you could always 
add more capacity to Abilene.

 But my main point here is to note that improvements at the transport 
layer are called for (if only to permit the 5 Gb/s flows to happen) and 
that those improvements should be driven (in part) by the need to share the 
net with jitter intolerant applications such as voip.

 Regards,
       -- Guy

--On Thursday, October 23, 2003 13:32:15 -0400 Jon Zeeff 
<> wrote:

> I'm not concerned with voip traffic filling Abilene.  But the days of "no
> packet loss" on Abilene are coming to an end.  We have researchers who
> can push 5 Gbits/sec. So a couple of them at the same time and you have
> Abilene packet loss/jitter/latency.




---------------------------------------------------------------wg-voip-+
For list utilities, archives, subscribe, unsubscribe, etc. please visit the
ListProc web interface at 

   http://archives.internet2.edu/

---------------------------------------------------------------wg-voip--