On Sat, 2005-08-13 at 12:37 +0700, Hay Tran The wrote:
> Hi Alex,
> Thanks for your answer.
>
> I am using Redhat Linux ES 3.1. gnuplot has been
> installed. ...
>
> I ran reporter as below:
> ./reporter --label "Test" /opt/polygraph/polygraph.log
> --plotter /usr/local/bin/gnuplot
>
> The error message is:
> plotter location must be specified
The --plotter option must precede the log filename:
./reporter --label "Test"
--plotter /usr/local/bin/gnuplot /opt/polygraph/polygraph.log
> Another questions:
> I uses simple test, req_rate = 1/sec, number of robot
> = 800, size = exp(13KB), I can gain around 677
> requests/second.
>
> If i change size = exp(80KB), I can gain around 119
> requests/seconds.
This is because it takes more time and resources to process larger
responses.
> For both cases, CPU used by Squid process ranging from
> 19% to 25%.
In this case, Squid CPU usage is not a good indicator because
(apparently) that is not the bottleneck. The bottleneck could be system
CPU usage, network bandwidth, disk bandwidth, etc.
> My question is why the number of requests/second drop
> down too much like this?
Probably because Squid cannot read and write data from/to disk that
fast, but there could be other bottlenecks as well.
> I want to know the best throughput of my cache server.
> With different test cases, I got different throughput.
> For the same test case, it seem to result to the same
> throuthput dispite the size of test object is
> different.
Not sure what you mean by "the same throughput" "for the same test case"
where [mean] object size is different. It is natural to get different
maximum throughput measurements depending on object size or other key
parameters. I would expect the same in real life if your workload
changes from small responses to large ones (e.g., the majority of users
is suddenly interested in sites with large embedded images or something
like that).
Alex.
This archive was generated by hypermail 2b29 : Mon Feb 06 2006 - 12:00:28 MST