Subject: Re: [linux-audio-dev] Distributed audio processing
From: John Lazzaro (lazzaro_AT_CS.Berkeley.EDU)
Date: Thu Oct 05 2000 - 20:49:58 EEST
> Theory apart, using current infrastructures (error correction modems,
> etc.), and following your experience, UDP has shown useful to obtain
> better latencies?
The zero-order issue is this: lets say you have a nominally low-latency
link (for example, Stanford-Berkeley round-trip times from Linux machine
to Linux machine is < 4ms now), but 1 in 100 packets get delayed to
20-30ms or dropped altogether (which is how the Stanford-Berkeley link
seems to operate during heavy-load time periods).
If you use TCP, that late or dropped packet means everything behind it
waits for retransmission and/or arrival -- since TCP is delivering a
reliable byte stream to the application. Whereas with UDP that has
media-aware packetization, in an application that can handle the
lost packet via the methods I described earlier, life goes on. So,
for a good-link and good-onramp situation in real life (this Stanford
Berkeley link is what I'm doing sfront internetworking development on,
along with a Caltech-Berkeley link to get a feel for longer speed-of-light
delays), that's a key way TCP falls down.
But what if you're not living in NIC <-> Internet2 <-> NIC land?
Here's a great webpage:
http://rescomp.stanford.edu/~cheshire/rants/LatencyResults.html
albeit it a bit out of date now, measuring latencies for xDSL,
cable modems, and normal modems in the real world; part of
Stuart Cheshire's article "It's the Latency, Stupid":
http://rescomp.stanford.edu/~cheshire/rants/Latency.html
For lots of brands of hardware, Cable Modems and xDSL are
relatively sane on-ramps: 5-8ms latencies seen in real life.
Dial-up modem users have no hope of seeing these sort of
single-digit latencies needed for tight-human-feedback-loop
apps, unfortunately, as Stuart Cheshire's explains in detail ...
--jl
-------------------------------------------------------------------------
John Lazzaro -- Research Specialist -- CS Division -- EECS -- UC Berkeley
lazzaro [at] cs [dot] berkeley [dot] edu www.cs.berkeley.edu/~lazzaro
-------------------------------------------------------------------------
This archive was generated by hypermail 2b28 : Thu Oct 05 2000 - 21:25:14 EEST