Re: PM3 slower than PM2

Christer Olsson (cox@clavicula.mednet.gu.se)
Sat, 26 Apr 1997 01:29:53 +0700 (MDT;)

On Thu, 24 Apr 1997, Stephen Fisher wrote:

> They (and you?) are basing quality of the PM3 on stupid ping times?
>
> The main problem is that the V.42Bis "dictionary" of compression rules is
> only 512K long in ComOS right now but very soon (if not already) they will
> have it upgraded to 2048K like most modems have. This should help the
> ping rates a lot.

Thats true for big ping packet yes. But what we're testing here is the
turn-around time for small packets. These are important for fast response
in game, online telnet and so. With 14400 bps modem and PM3, I'm getting
190 ms as best, and that's too much because I'm getting only 135 ms with
USR Sportsters and the old Linux-based modemserver.

The problem PM3 seems to have is that it's waiting too long time for more
data before it sends the packet over the line. Because the PM3 doesn't
have any slow serial line between the ethernet-side and the modemcard,
the ping-time should be lesser, more like under 100 ms or so against mine
14400-modem (connected with 57600 bps).

What I'm missing in PM3 is the dynamically timeout for receiving data
before sending the packet. PM3 is waiting too long time even with low
load.

The dictionary size is very important for compressible data (the values
is 512 bytes and 2 Kbytes, not 512k and 2048K). I've tried myself with
WWW and News. Most WWW-data sendning in the beginning is text-data (who
is sent before the already compressed pictures) and a large dictionary
table gives a significant speedup for the text-part in WWW. If a page
need 3 seconds with 512 bytes for updating the text-parts, it only needs
2 seconds with 2048 bytes dictionarys and you'll easily note that.