>>>>>>>>>>>>>>>>>> Original Message <<<<<<<<<<<<<<<<<<
On 12/5/00, 1:42:41 AM, [EMAIL PROTECTED] wrote regarding Re:
Benchmarks / Performance Tuning:
> It is very important to use "ab -c 20 " ( or more ) to get real
> information about how the server works under load - "ab -c 1 " is of no
> real use.
Unless, of course, you'd like to estimate the average CPU demand on your
machine, with your VM etc. by firet starting up sar and sa. Why might
this info be useful?
If you know that the average CPU demand to process a request is 3 ms, (when
C = 1) you can infer that the maximum average throughput will be 1000/3 =
333 requests per second: you don't need to do any ab tests to find out what
the maximum average is (unless, as with apache, cpu usage goes down as C
goes up). If you do as Costin suggested, you might find that the actual
average throughput with C > 20 will be less than the maximum.
Of course, knowing the avg CPU demand and the maximum average throughput
don't tell you how/where to tune, but it might give you a sense of
whether tuning is even necessary.
An example of how this approach as applied to Apache can be seen at
http://members.bellatlantic.net/~designrw/BENCH1.zip
I did some preliminary measurements fot TC4.0-m4 and plan to do some more
after the next milestone (plus use some of the other tools mentioned by
Costin).
Roy
> Costin
