Re: [XCON] On encodings - observations

[Adam Roach <adam at nostrum.com>]

[not as chair]

Henning Schulzrinne wrote:

> Right. Mobile phones have processors running at several hundred MHz, 
> i.e., roughly equivalent to desktop processors in the late 90s. I 
> don't think they'll break a sweat encoding 200 bytes in gzip.


Your assertion is not completely well founded: many of the phones sold 
today are running in the tens of MHz, not hundreds. Even for smart 
phones, it's not unusual to get ahold of terminals based on 33 MHz 
DragonBall VZ processors. (If you like, I'll ask a few questions during 
the XCON meeting next week and probably find a dozen or so people in the 
meeting actually in possession of one). I have run tests on these 
specific processors using Deflate and 1000-byte messages. The resultant 
compression time is on the order of 300 - 500 ms. For smaller messages 
-- say, 200 to 300 bytes or so -- you're looking at about 100 ms to 
compress such messages.

For messages in that size range, you'd be hard pressed to make up 100ms 
of latency by sending fewer bits.

When I did my analysis of this topic, my conclusions were: on processors 
at this speed, and using the network data speeds we discussed before, 
using Deflate compression is pretty much a wash (in terms of latency) 
for messages of any size. The amount of time spent compressing a message 
was always nearly equal the amount of time you saved in transmission. 
There are some much lighter-weight, *patented* compression algorithms 
(e.g. v.44) that are orders of magnitude faster, but I'm not certain we 
want to mandate a patented compression algorithm for all users of the 
conference control protocol.

/a

_______________________________________________
XCON mailing list
XCON at ietf.org
https://www1.ietf.org/mailman/listinfo/xcon