Re: Biiig machines with pmix3-ips.pl
Alex Rousskov <[email protected]>
| Newsgroups | gmane.comp.web.web-polygraph.user |
|---|---|
| Message-ID | <Pine.BSF.4.10.10011291347250.58951-100000@measurement-factory.com> |
On Wed, 29 Nov 2000, Thomas Proell wrote: > I want to run polygraph on (very?) big machines (dual P3 / 512MB). > Unfortunately I don't have many of them, but I want to generate as > much realistic load as possible. > I'm about to set up the alias-IP-addresses, but the pmix3-ips.pl > tells me to use only a maximum of 400 robots on one client. > > Why? Is this recommended, or what's up? Even on a 80386 it > will tell me to use 400 robots. The number of aliases/robots is 1000. 400 is the load (req/sec). These per-host limits are required by PolyMix-2,3 rules to make the results comparable and reproduceable. The actual values were selected so that an average PC can handle the load even if a proxy gets behind from time to time. No interesting PolyMix variables depend on your host capabilities. This is probably a good thing because it makes comparisons easier. Custom workloads do not have to follow PolyMix rules, of course (see below). > Isn't there a better aproximation then that? Approximation of what? > Can I simply change the constant to another value? What value > would you suggest? You sure can violate PolyMix-3 requirements and use as many robots per machine as your hardware and OS can support. The only "real" difference that I can see is in the number of MAC addresses and ports. Whether that difference is important is impossible to say in general. Note that using more than 2000-2500 robots on fast Ethernet is probably not a good idea as you will hit the network bandwidth limit. To estimate the maximum number of robots per host, I would suggest to run a [relatively short] no-proxy test and compare performance (response time) in idle and top phases. The response time change should be negligible. Alex.