El 23/04/2013 22:32, "Albert Cervera i Areny" <[email protected]> va escriure: > > A Dilluns, 22 d'abril de 2013 11:24:15, Udo Spallek va escriure: > > > We could try a test script and measuring the pystones. > > > It is a whetstone[1] implementation for python and found in the stdlib > > > *test*. > > > The idea is to run a fixed script on an individual computer hardware and > > > measures the run time and a ratio compared to a reference computer: > > > >>> from test import pystone > > > >>> time_running, mypystones = pystone.pystones() > > > >>> mypystones > > > 125000.0 > > > My computer has 125000 pystones or 125 kpystones. > > > If I have a test which takes 1.4 seconds to run on my hardware, I can > > > calculate the kilo pystones for the test: > > > >>> kpystones = mypystones * 1.4 / 1000 > > > >>> kpystones > > > 175.0 > > > The test with 175 kpystones are comparable on different computers - in > > theory. > > > On the other hand is the performance of the Tryton server affected by > > > many different technologies, not only cpython. > > > So I don't know if we have a chance to compare the timings of the test > > > results at all. > > I don't think that works. We're talking about CPU, RAM, HD Speed, running applications, etc, etc.
I think it could be interesting to have the time of execution of the diferent tests and collect them in a wiki page with the configuration of the server where the test has veeb executed. It could be a table with the diferent cases as columns and the configuration and test times in rows. Configurqtions are CPU, ram, cache but also version of postgres and other implied technologies and some params in their configurations. It's impossible to take in account all variables but ut could help to identify what are the most important. Guillem Barba
