Daniel Stenberg wrote:

> I figure you just run a test with libcurl compiled as before, then
build a new version with increased CURL_MAX_WRITE_SIZE and re-run the
same test. Do you spot any difference?

I have some figures for you: the following table shows transfer rate
test results for various situations. All tests have been performed with
100Mbytes transfer on a 100M ethernet LAN, and rerun 3 times to get a
reasonable average.
I have to say that tests made against an OS400 server were not very
precise: I have noted ~20% drift between some of the 3 (same) tests in
this condition (probably due to some file and/or system overhead).
Client on OS400 has a very low drift.

Client direction server protocol kbytes/sec_at_16384-byte buffer
kbytes/sec_at_40960-byte buffer
------ --------- ------ -------- ----------------------------
----------------------------
XP(cygwin) < linux ftp 11430 11381
XP > linux ftp 8349 9456
XP < os400 ftp 7533 7694
XP > os400 ftp 4911 (+/- 40% !) 5649
Linux < os400 ftp 7539 7616
Linux > os400 ftp 9100 9100
Linux < os400 http 7304 7130
Linux > os400 http 7167 7224
XP < linux http 11514 11516
XP > linux http 8906 10269
XP < os400 http 7694 7384
XP > os400 http 3898 4235
Os400 > linux http 8344 7495
Os400 < linux http 7232 7312
Os400 > linux ftp 7988 7018
Os400 < linux ftp 7146 7204

As a conclusion, I do not feel very excited about raising the buffer
size! In some cases, the results are even worse.

I hope it helps
Regards,
Patrick
Received on 2009-08-26