On Mon, 5 Mar 2001, Artem Veremey wrote:
> I haven't noticed that polymix workloads have rate set up for
> clients.
A benchmark is pretty useless if you cannot tell it to generate a
given request rate. Best-effort workloads (i.e., those that go as fast
as the device under test lets them) should not be used for performance
analysis.
Use Robot's req_rate field to specify a request rate for "constant
request rate" workloads:
http://polygraph.ircache.net/doc/pgl/types.html#Pgl:Robot
> What is the maximum request rate you can get out of one polyclt?
It depends on the workload, OS, and hardware. We can get 700+ req/sec
using PolyMix-3 on 450 MHz P3 machine running FreeBSD. A completely
different workload (with zero object sizes) in similar environment
yields more than 1700 req/sec.
Whatever your maximum load is, make sure that your setup can produce
more than what the DUT can handle so that you know the DUT is the
bottleneck.
> Web page says that polyclt running on FreeBSD can generate up to
> 700h/s. In my case, I can't get more then 150-180h/s out of
> polyclt running modification of simple.pg on Solaris. Is such a
> huge difference expected?
IIRC, simple.pg uses one best-effort robot. To get higher traffic
rates you need to increase the number of best-effort robots or,
better, use "constant request rate" robots. In general, you should not
use simple.pg for anything but the simplest "connectivity" tests. Use
PolyMix, WebAxe, or your own custom workload based on the standard
ones.
Alex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:18 MDT