From Documentation/cgroups/freezer-subsystem.txt, the following words describe the purpose of cgroups freezer: "The cgroup freezer is useful to batch job management system which start and stop sets of tasks in order to schedule the resources of a machine according to the desires of a system administrator." IMHO, when system load is high, administrator could freeze some of cpu-hogged tasks in favor of more important ones, once system is most likely to be idle, frozen tasks could be thaw back on line. I don't know this idea is right or wrong, can somebody give me some hints about practical usage of cgroup freezer? thanks -- I am a slow learner but I will keep trying to fight for my dreams! --bill
Hi Bill :) On Mon, Feb 20, 2012 at 16:50, bill4carson <bill4carson@gmail.com> wrote:
IMHO, when system load is high, administrator could freeze some of cpu-hogged tasks in favor of more important ones, once system is most likely to be idle, frozen tasks could be thaw back on line.
what you describe, IIRC, is something doable via "batch" command
I don't know this idea is right or wrong, can somebody give me some hints about practical usage of cgroup freezer?
IMHO, it's also useful if you manage computer grid. So say you have 16 nodes. A job runs on node A, but later you found that node C has the least load. You freeze the job and then unfreeze it in C. Of course, it assumes that the nodes are homogen. And yes, it sounds very similar to virtual machine migration. basically, both of them have same principle on how to get things done. -- regards, Mulyadi Santosa Freelance Linux trainer and consultant blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com
On 2012年02月20日 23:06, Mulyadi Santosa wrote:
Hi Bill :)
On Mon, Feb 20, 2012 at 16:50, bill4carson<bill4carson@gmail.com> wrote:
IMHO, when system load is high, administrator could freeze some of cpu-hogged tasks in favor of more important ones, once system is most likely to be idle, frozen tasks could be thaw back on line.
what you describe, IIRC, is something doable via "batch" command
I don't know this idea is right or wrong, can somebody give me some hints about practical usage of cgroup freezer?
IMHO, it's also useful if you manage computer grid. So say you have 16 nodes. A job runs on node A, but later you found that node C has the least load. You freeze the job and then unfreeze it in C.
Thanks for your reply :) Just one silly question, how could frozen task be moved from node A to C ?
Of course, it assumes that the nodes are homogen. And yes, it sounds very similar to virtual machine migration. basically, both of them have same principle on how to get things done.
Yes, KVM+qemu have already support this feature. -- I am a slow learner but I will keep trying to fight for my dreams! --bill
Hi Bill :) On Tue, Feb 21, 2012 at 08:23, bill4carson <bill4carson@gmail.com> wrote:
Just one silly question, how could frozen task be moved from node A to C ?
I forgot to mention that I haven't tried the freezer feature in Linux kernel, but I got impression that is quite similar with so called checkpointing. So assuming that it's indeed checkpointing, what you need to do is just grab the checkpointed image, copy to another node and execute a command to unfreeze it. Usually, the image could be similar with ELF dump core, but it could be something else too. -- regards, Mulyadi Santosa Freelance Linux trainer and consultant blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com
On 2012年02月22日 01:08, Mulyadi Santosa wrote:
Hi Bill :)
On Tue, Feb 21, 2012 at 08:23, bill4carson<bill4carson@gmail.com> wrote:
Just one silly question, how could frozen task be moved from node A to C ?
I forgot to mention that I haven't tried the freezer feature in Linux kernel, but I got impression that is quite similar with so called checkpointing.
So assuming that it's indeed checkpointing, what you need to do is just grab the checkpointed image, copy to another node and execute a command to unfreeze it. Usually, the image could be similar with ELF dump core, but it could be something else too.
Thanks for your information, I'll make some time to try what checkpoint does. -- I am a slow learner but I will keep trying to fight for my dreams! --bill
participants (2)
-
bill4carson -
Mulyadi Santosa