Hi, This is a follow up to the early post of "Power aware credit scheduler" http://lists.xensource.com/archives/html/xen-devel/2008-06/msg00580.html. We are now considering implementing some of the points. The implementation candidates are: a.1) csched_cpu_pick: when picking pcpu for vcpu migration, take package C state into consideration, and pick idle core with busy sibling already. a.2) csched_vcpu_wake: current implementation will blindly tickle all the idle pcpus for grabbing the waked vcpu. If take power into consideration, it may be better to only tickle one idle pcpu. The a.1 is quite straight forward, and considering it may have performance impact, we will add xen option to enable/disable it. For the a.2, we are not sure and are looking forward comment from community. The major concern is: if tickling only one idle pcpu, will it impact the performance? If so, how much the impact would be. One scenario I can think of is that: after tickling the idle pcpu, the pcpu may unfortunately become busy and have no chance to grab the waked vcpu, so the waked vcpu has to wait for the next scheduler SOFTIRQ which make I/O latency longer than expected. But how much possibility would that happen? Comments is welcome, Especially for the a.2, is it acceptable to tickling only one pcpu? Best Regards Ke _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel