[Somehow Gnus mimified the message wrong (not the first time), so that
the initial text was invisible for those with MIME-aware mail readers.
This is the text that was supposed to go out with the message.]
I'm planing on running a "Web Client Library BOF" at the ORA Conference
this year (have a timeslot on Thursday;
http://conferences.oreillynet.com/cs/os2001/pub/10/bofs.html).
As a bit of research before the BOF I today tried out various
libraries that are available to Perl. One of the things I tried was
to write a little loop that fetches the ActiveState homepage 50 times.
I got kind of strange results:
LWP 1.8s 0.45s 0.07s 5680 kB
LWP::Simple 1.8s 0.15s 0.07s 4432 kB
HTTP::GHTTP 10.9s 0.06s 0.07s 3484 kB
HTTP::Lite 11.2s 1.56s 0.13s 5244 kB
HTTP::Webdav 1.3s 0.01s 0.02s 4152 kB
The first number is real time, the second is user CPU time and the
third system CPU time. The last number is the size of the process
when the loops finish.
What I don't understand is why HTTP::GHTTP and HTTP::Lite get so bad
real time numbers. Can anybody else reproduce similar results or
explain it? I have attached my test scripts.
The HTTP::GHTTP manpage site a benchmark that says:
Benchmark: timing 1000 iterations of ghttp, lite, lwp...
ghttp: 8 wallclock secs ( 0.96 usr + 1.16 sys = 2.12 CPU)
lite: 21 wallclock secs ( 3.00 usr + 3.44 sys = 6.44 CPU)
lwp: 18 wallclock secs ( 9.76 usr + 1.59 sys = 11.35 CPU)
Is that benchmark script available somewhere so that I can verify its results?
Regards,
Gisle
