On Thu, 27 Jul 2000, Derek Johnstone wrote:
> I've checked out what you said about the full cache, and it turns out to be a
> rather unhelpful coincidence, whereby the cache disks are full after 4 hours
> at that particular request rate. So that would explain the results seen, and
> thanks for that.
Please note that you are supposed to run any performance test, including
DataComm, with the cache pre-filled. We recommend filling the cache with
the low-hit-ratio workload that is otherwise similar to the performance
workload. The fill goal should be specified to be twice the cache
capacity. Polygraph does not re-use objects from the fill run in the
performance run so your hit ratio will not be affected. The performance
of a cache may be affected though; many caches perform better when they
are "empty".
> The Throughput from the executive summary has been showing rates
> greater than 50% higher than that specified by Req_Rate in the
> datacomm file. You are correct that the "Engineering Summary" shows
> offered and measured load to be the same, but it is this higher than
> expected value. (E.g. Req_Rate=200req/sec gave a Throughput of
> 331.16rep/sec in the Executive Summary.). Is this normal? What am
> I seeing here?
This is not normal. If you want, you can send me a report along with the
console output from client and server side(s) (all in a compressed tar
file), off the list (or send me a URL if you prefer). I will try to look
at the report and logs, but may not be able to do it promptly.
Alex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:14 MDT