On Thu, Sep 21, 2006 at 10:08:26AM -0700, Matt Anger (manger) wrote: > Following the vsched explained page I am trying to setup a case to > adjust the priorities of my vservers. > EX: 4 vservers at contexts > 49186 > 49187 > 49188 > 49189
first, do not use dynamic contexts, they are deprecated since a while now and will be removed shortly, use static context ids instead ... > I want the distribution of CPU time to be 40,20,20,20 > So I use vsched and do > vsched --xid 49186 --filrate 2 --interval 5 > vsched --xid 49187 --filrate 1 --interval 5 > vsched --xid 49188 --filrate 1 --interval 5 > vsched --xid 49189 --filrate 1 --interval 5 > > Then I start a CPU intensive process (simple C program that measures > how long it took a computer to do a large loop of arithmetic, usually > takes about 100s on the P3 800 I am doing tests) in all 4 vservers. > Unfortunately it still schedules them fair-share according to vtop > and according to the program itself (All programs report 400 s for > completion). I also tried playing with tokens, tokens-min, tokens-max, > but no matter what I did it still scheduled them fair share. Am I > doing something wrong? probably, you want to make sure that the following requirements are met: - CONFIG_VSERVER_HARDCPU is enabled in the kernel - you have set the sched_hard flag for that context - the max_tokens value is relatively low (otherwise the process will get a boost at the beginning) - the 400s are wall time, not process time if you can confirm all of them, please contact me at the irc channel, as this would mean that there is a bug somewhere ... HTH, Herbert > 2.6.17.11-vs2.1.1-rc31 > util-vs 0.30.210 > > Matt > _______________________________________________ > Vserver mailing list > [email protected] > http://list.linux-vserver.org/mailman/listinfo/vserver _______________________________________________ Vserver mailing list [email protected] http://list.linux-vserver.org/mailman/listinfo/vserver
