2011/12/6 Arie Skliarouk <sklia...@gmail.com>: > Hi, > > I understand that this is not the quite appropriate mailing list to ask the > question, but the question is related to the LXC tech we use on the server, > so here it goes: > > Most of the time the LXC containers on our servers work properly, but > occasionally someone, somewhere starts an IO heavy operation that kills > performance for everybody. For some time I tried to ask people nicely to use > "ionice -c 3" or run the task offhours but this is not enough. The problem > happens quite often for people to complain, but not (IMHO) to warrant > purchasing of new hardware. > > I envision that an ideal solution would be some daemon that would monitor > disk IO activity and automatically reduce (or raise, depending how you view > it) ionice priority of the process or the container. The daemon would > restore the IO niceness after some good behavior period. > > Is there any solution along the lines? > > -- > Arie
Hi, Basically cgroup's blkio controller works for such scenario. Please see http://www.mjmwired.net/kernel/Documentation/cgroups/blkio-controller.txt for details. In my test I found that the blkio controller shipped with older kernel (e.g. rhel6's 2.6.32) doesn't work quite well when the all IO workers in each group are very seeky ones, unless you echo 0 > slice_idle to switch to iops mode manually. However in the latest upstream kernel it works very well by the default arguments, without any adjustment. Thanks, Zhu Yanhai ------------------------------------------------------------------------------ Cloud Services Checklist: Pricing and Packaging Optimization This white paper is intended to serve as a reference, checklist and point of discussion for anyone considering optimizing the pricing and packaging model of a cloud services business. Read Now! http://www.accelacomm.com/jaw/sfnl/114/51491232/ _______________________________________________ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users
