On Fri, 13 Apr 2001, Shuowen Yang wrote:
> I found that hit rate drops with time considrably when simple
> workload is used. Our system has a very large disk to
> cache objects, no swapping happened. Polygraph did not
> show any error message or warning in the test period.
> Testing period : 20 min
> Req_rate : 20 /s
> DHR: 55%
> In the first five minute, hit rate is around 55%, at the end of
> the test hit rate is around 22%.
> Any suggestions. Thanks.
I would start with:
- Generate a report to check that _offered_ hit ratio
stays at 55% throughout the test. See User Manual.
- Check proxy logs for possible reasons why the objects
are not getting cached OR why old objects get removed.
- Start a second (concurrent!) Polygraph test with low request
rate after the hit ratio of the first test has dropped.
See if the second test yield good HR. This will tell you
whether its old or new objects that are not producing hits.
- Enable "--prn_false_misses" option in polyclt to analyse
false misses (this is a time consuming step, try to avoid
it if you can, see option description on the Web).
If you collect more facts, but are still stuck with the problem,
please send your new findings to us, and we will go from there.
In general, please note that it is a bad idea to use simple.pg for any
kind of benchmarking other than simple "connectivity" tests.
HTH,
ALex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:18 MDT