Juan Villar wrote: > Christopher Friesen escribió: >> Does anyone have any results for the performance of the latest ckrm >> cpu controller? >> >> I'm interested in scenarios with multiple classes in a hierarchy with >> varying shares assigned to each class, with one or more cpu hogs in >> each class. >> >> I'm curious to see how well it does in terms of meeting the share >> targets. >> >> Chris >> >> >> _______________________________________________ >> ckrm-tech mailing list >> https://lists.sourceforge.net/lists/listinfo/ckrm-tech >> >> > Hi, > > I'm also trying to measure the performance of cpu controller and how it > shares the > resources and how it meets the shares so that I set up a system with > CKRM f0.8 > and CPU controller v0.4 in a kernel 2.6.17-rc3. The machine is a Pentium > III. > > My test was based on > http://ckrm.sourceforge.net/downloads/ckrm-linuxtag04-paper.pdf > with same differences. > I define four groups (GOLD,SILVER,BRONZE,BEST EFFORT) with shares of > (60,30,9,1). > Afterwards, I populate each group with five tasks with different > priority levels > (-20,-10,0,10,19). These task all execute the same code: > > counter=0; > while (1) { > sqrt (rand()); > if (++counter>=1000) { > printf("Loop\n"); > counter=0; > } > } > > Moreover, each task writes to a different file and I count the printed > lines in these > files after 30 minutes of execution. > > For example, this is a figure with the results. As you can see, the > number of printed > lines of tasks with priority levels (0,10,19) are similar between > groups, however tasks > with priority levels of (-20,-10) are more different. I think it is due > to the difference > between timeslice values, but I think there is not enough difference > from a share point of > view on the whole. > any suggestion?
Current implementation, interactive tasks get extra bonus, so perhaps these tasks are regarded as interactive tasks. If that is the reason, increasing the number of loop may mitigate the problem. > Nice 250 Hz Nice 250 Hz > ----------------------- > -20 800 0 100 > -19 780 1 95 > -18 760 2 90 > -17 740 3 85 > -16 720 4 80 > -15 700 5 75 > -14 680 6 70 > -13 660 7 65 > -12 640 8 60 > -11 620 9 55 > -10 600 10 50 > -9 580 11 45 > -8 560 12 40 > -7 540 13 35 > -6 520 14 30 > -5 500 15 25 > -4 480 16 20 > -3 460 17 15 > -2 440 18 10 > -1 420 19 5 > > The scheduler assigns the same penalty for every task (+5) so the > dinamic priority are > proportional to nice values (-15,-5,5,15,19). > > juanan > > ------------------------------------------------------------------------ > > > ------------------------------------------------------------------------ > > _______________________________________________ > ckrm-tech mailing list > https://lists.sourceforge.net/lists/listinfo/ckrm-tech Thanks, MAEDA Naoaki _______________________________________________ ckrm-tech mailing list https://lists.sourceforge.net/lists/listinfo/ckrm-tech
