Re: [httpstreaming] [dispatch] [conex] Q-HTTP

[Mikael Abrahamsson <[e-mail]>]

On Fri, 12 Nov 2010, DIAZ VIZCAINO, LUIS MIGUEL (LUIS MIGUEL) wrote:

> And testing is just a MUST. We need to know what is going wrong for THAT 
> flow, because not all the applications/flows have the same requirements.

No, testing is not a MUST. If you need to know what is going on with THAT 
flow, then you make sure you have insight into what THAT flow is 
experiencing, you don't inject more traffic.

Your time would be much better spent developing an API into the client IP 
stack to give insight into what TCP/RTP is actually experiencing 
(timestamping used to deduce jitter etc) than to create *more* traffic 
when facing adverse conditions.

And since you say "flow". I run *routers*. Routers act on a per-packet 
basis. They don't act on flows. They act on precedence/dscp values in each 
packet and queue according to policy if it exists, otherwise it does FIFO.

I hate testing. I'd much rather have insight into real traffic than 
creating test traffic. Test traffic can look fine where the real traffic 
is not. It's just amazing that we're sitting here some 30 years after TCP 
was invented and still have no tools for users to see what TCP is doing, 
short of dumping the packets on their interface and running wireshark on 
it.

-- 
Mikael Abrahamsson    email: [e-mail]