[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[ns] web caching simulations -- bug in FullTcp?




I'm having trouble simulating a web cache hierarchy under a system under
moderate load. If I run simple-webcache-trace.tcl with tracefiles that I
generate, most requests are satisfied quickly, but others take a very long
time to get satisfied (up to 30 seconds). This seems to happen whenever
the load exceeds an average of 5 requests per second, exponentially
distributed. I've tried (1) distributing this load across many Http
clients and also tried (2) calling http-agent's send-request function
directly, to no avail.

It looks like the request is being sent (line 239, http-agent.tcl), but
for some reason the cache doesn't execute the callback function (line
162, http-cache.tcl) for a very long time. If I force my tracefile
generator to leave a gap between requests equal to the propagation delay
of the link, things work fine. Further, if I change line 26 of
http-agent.tcl to use SimpleTcp instead of FullTcp, things work fine. This
problem appears to be independant the topology and pagepool type.

I emailed Haobo and he believes he has encountered this problem before. I
was wondering if anyone has experienced similar problems. If so, have
you found a workaround? Does anyone have any ideas about what might be
wrong? If anyone has solved this problem, would you mind sharing your
code?

Any help would be greatly appreciated.

Thank you,
Matt