Internal modem / FIFO buffer / throughput test
Franc Zabkar
Rockwell ACF2 chipset. Recently I was curious as to what effect, if
any, a variation of the Tx/Rx FIFO buffer trigger points would have on
throughput, and on overrun errors. For the following tests I used
WS_FTP (ver 5.06) to upload and download a highly compressible file to
and from my webspace. This file consisted of the letter "A" repeated
one million times. The transfer rates are as reported by WS_FTP.
Here are my throughput results (KBps, Win95B):
Tx/Rx FIFO buffer trigger levels
disabled min 2nd notch 3rd notch max
upload 19.40K 16.53K 19.40K 19.37K 19.52K
download 12.45K 15.63K 15.62K 15.62K 15.62K
The port rate for these tests was 230400bps. Windows System Monitor
indicated that the instantaneous throughput rates never exceeded
22KBps, suggesting that the transfer speeds were limited by my
hardware (Cyrix MII-333, ISA bus) and/or software. My ppplog,
modemlog, and modem post-call diagnostics recorded no overrun errors
during any of the tests. Each test was repeated several times over
several days and was reproducible to within +/- 0.1 KBps.
One curious result is the low upload rate of 16.53KBps for the min
setting of the Tx FIFO. This is less than the throughput when the FIFO
is disabled.
Another odd result is the difference in download throughput between a
disabled FIFO, and one set to the lowest notch. As the actual Rx
trigger points correspond to 1, 4, 8, and 14 bytes, I would have
thought that a minimum setting of 1 would produce the same results as
for a disabled FIFO. Perhaps I'm misunderstanding something here?
My final observation is in relation to the low download speed of
12.45KBps in the case of a disabled FIFO. I believe this reduced speed
is due to Win95's interrupt latency and overhead, not to any
deficiency in the modem.
-- Franc Zabkar
Please remove one 'g' from my address when replying by email.