Hello All,
I want to test how file system performance can affect Squid Web Cache
performance. I decided to test the following file systems: ext3, ext2
(with options: noatime, async), reiserfs and xfs. Problem is that I don't
know how to write appropriate workload. I want polyclt robots to increase
gradually request rate until DUT achieve maximum of acceptable value of
response time. Then polyclt needs to adjusts local population of robots in
order to sustain an acceptable value of response time and finally fill the
cache twice.
I've spend some time on writing a workload but it isn't so easy. My first
approach was using Rptmstat but I probably did some mistakes and rptmstat
didn't work for me. Finally I decided to try a watchdog feature. So, I
simply copied some parts of standard polymix-4.pg and "include" files and
build below workload file.
I've ran a short test on Squid Web Cache Proxy. I've used one 4,3GB IDE
disk for a cache cause I've had no time and equipment to make a productive
test. However, test like that should shows that file system have influence
on Proxy performance. Unfortunately my results don't confirm above
statement.
I'm not confident if I've written workload properly. I wanted to write a
constant response time workload. I've used two watchdog to control number
of robots alive. Every 2000 xactions "upSample" watchdog increase populus
factor by +1% if mean response time is below 300msec. Second watchdog,
"downSample", every 1000 xactions decrease populus factor by -1% if mean
response time exceeds value of 400msec. Nothing is done if mean response
time is between 300msec and 400msec. During the test I observed numerous
response time spikes. I'm not sure if I selected a valid response time
window. Moreover I remained throughput as an input parameter which I think
now is not good idea. In addition to this I set low recurrence in a robot
section. It seems to me logical that lower hit ratio then greater number
of disk I/O operations. Perhaps hit ratio should be equal to zero.
If someone has tested mentioned above scenarios please send me workloads
or tell me where I've made mistakes.
Thanks in advance,
Peter
This archive was generated by hypermail 2b29 : Mon Feb 06 2006 - 12:00:28 MST
