On Mon, 12 Aug 2002, Ko Tensei@KIT wrote:
> A real cache may take 7 days to be 'filled', while Polygraph takes
> only several hours by using shortcuts to "compress time".
>
> However, in terms of PolyMix-4, waiting 12 hours is still a little
> long for research (such as algorithm tuning) purpose.
>
> Is it possible to reduce the PolyMix-4 test time to the degree of top2 phase
> (say 4 hours~)? Any suggestion is welcomed.
Yes, it is possible. You can alter PolyMix phase schedule any way you
want (the schedule is defined in polymix-4-guts.pg file). Any
significant change will have side-effects. Whether those side effects
are important to your research would depend on your device under test
and the algorithms you want to stress.
> Would the following steps work, when using Web Polygrpah?
> 1. Propotionally reduce cache_size configuration in proxy server machine
> (hd and mem),
> and reduce obj_size received from polysrv (content-type table tuning).
Yes, you can create a workload with smaller objects. This will help
you to increase request rate. However, the overheads of disk I/O may
be a lot different. I did not know whether you care about that.
> 2. On the other hand, any hard-coded 'definition's in proxy server source
> which distiguish 'small' object with 'large' object might also be tuned.
> Of course, as to those friendly proxy servers, configure file tuning
> should be sufficient.
While you may be able to tune your proxy, the underlying OS and file
system code (if any) may have its own optimizations and assumptions.
Again, it is impossible to avoid all side-effects in general. You
should assess the expected effect on the algorithms you study and
ignore other, irrelevant side effects. In other words, is shorter test
with smaller objects going to put your proxy in a steady state, as far
your performance characteristics of interest are concerned?
Alex.
--
| HTTP performance - Web Polygraph benchmark
www.measurement-factory.com | HTTP compliance+ - Co-Advisor test suite
| all of the above - PolyBox appliance
This archive was generated by hypermail 2b29 : Mon Feb 06 2006 - 12:00:24 MST