On Fri, 6 Jul 2001, [iso-8859-1] sophia antip wrote:
> 1.even though polyct ends normally after completion of
> the 10 phases, I couldn't run lx and lr on the log
> files (no matching phases found). I understood that if
> polyclt ends correctly, these 2 commamnds must run. I
> don't get why they don't.
I think you forgot to tell Polygraph to log statistics using the --log
option:
http://www.web-polygraph.org/docs/reference/options.html#option:docs/reference/options/log
The --console file only captures console output. One must have a
binary log to use with lx/ltrace.
> 007.38| Connection.cc:206: error: 54/1458 (s10048)
> Only one usage of each socket address
> (protocol/network address/port)
> is normally permitted.
My guess is that this error message is an indication of a race
condition inside MS kernel when the kernel allocates a port number for
the new connection. IIRC, we have seen similar errors with Unix
kernels under heavy load. Try using --ports option to tell Polygraph
to manage port allocation internally instead of relying on the kernel
stack to do the right thing. We usually use --ports 3000:30000.
http://www.web-polygraph.org/docs/reference/options.html#option:docs/reference/options/ports
http://www.measurement-factory.com/docs/PolyMix-3/
> 008.58| Connection.cc:206: error: 720/4508 (s10061) No
> connection could be made because the target machine
> actively refused it.
> 008.58| error writing to 192.168.1.140:3128 after 0
> reads, 0 writes, 1 xacts
I suspect that whatever runs on 192.168.1.140:3128 refuses _some_ of
your connections. Perhaps the proxy is overloaded or, if it is a
software solution, the corresponding OS is overloaded. Make sure your
no-proxy test runs fine first so that you can confirm that the problem
is with the proxy.
You can also try to telnet to the proxy address when Polygraph test is
running and see if your connection is always accepted (not a very
reliable test, but might be useful).
> 008.68| Connection.cc:98: error: 3759/4745 (s24) The
> program issued a command but the command length is
> incorrect.
This error message does not make sense in the Connection.cc:98
context. The error is returned by the socket(2) system call that has
no "command" or command length argument. Ignore it for now, I guess.
> I'll like to know if they are the fact of
> misconfiguration or something else. Find the file
> attached.
My guess is that you are overloading the proxy and/or the Polygraph
machines. While the load levels you are using are not very high, you
still may need to tune your OS to handle them. For example, we have
to reduce MSL value to 3 seconds on Unix systems:
http://www.measurement-factory.com/docs/FreeBSD/
I do not know how to tune Windows though.
I would suggest to run a no-proxy test first (same workload, no
--proxy option) to eliminate the proxy as a suspect.
Alex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:20 MDT