Marc E. Fiuczynski wrote:
Hubertus,
A PlanetLab node typically has about a dozen or so active vservers (aka slices), but there are typically ~150 vservers put onto a such a node.
This means that I will have to likely create 150 or so entries in the /rcfs/taskclass directory. Presumably your scheduler only considers "active classes" i.e., those which have running processes, rather than all classes, right?!
During the scheduling process .. yes it only considers active classes and scheduling decisions are O(1) just like today's scheduler.
The only difference is that we associate today's scheduler queues with
each CLASS and then run an O(1) class scheduler on top of that.
So we obtain O(1) to determine the next class to run and O(1) the task in that class -> O(2) :-)
Just like today the O(1) scheduler is dependent on the number of active tasks not numtasks .. same applies here.
During aperiodic events such as load balancing etc. we do run through all classes and update a few metrices. Nothing earth shaking ... Similar as today where the rebalancing code has to run through all CPU's.
Any way, those are the round numbers for the typical use on PlanetLab which CKRM's cpu must support.
------------------------------------------------------- This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170 Project Admins to receive an Apple iPod Mini FREE for your judgement on who ports your project to Linux PPC the best. Sponsored by IBM. Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php _______________________________________________ ckrm-tech mailing list https://lists.sourceforge.net/lists/listinfo/ckrm-tech
