On Tue, 2006-08-22 at 11:44 -0700, Paul Menage wrote: > On 8/22/06, Chandra Seetharaman <[EMAIL PROTECTED]> wrote: > > On Mon, 2006-08-21 at 16:39 -0700, Paul Menage wrote: > > > On 8/7/06, Chandra Seetharaman <[EMAIL PROTECTED]> wrote: > > > > Hi Paul, > > > > > > > > I liked the basic idea of accounting for cpu cycles. But, i see it > > > > similar to the delay accounting feature that Vatsa pointed to than to be > > > > part of Resource Groups, as the main purpose of RG is resource > > > > management than resource monitoring. > > > > > > To my mind, those are intimately related - you can't do resource > > > management without resource monitoring, after all. If it would help > > > > That was the thinking we had originally and that is why e-series CKRM > > had delay accounting functionality. > > > > But later felt that it will be valuable to have the delaystats > > functionality (monitor _only_ functionality) separated from CKRM so that > > users can use it independently as well as with CKRM. That is the same > > rationale behind my suggestion that this functionality to be separate > > from CKRM/RG itself. > > But if you want to be able to monitor the resource usage of a group of > processes, including new/short-lived processes, surely you need some > kind of container abstraction to automatically group processes? Maybe
delay stats does that with tid. > we should separate out the container abstraction from RG, so that code > other than RG can use them too. (This would be similar to PAGG, I > guess). > > > > > Basically, the user space app forms the group and associates it with an > > > > RG, sums up delay stats for each task in the group to get the collective > > > > delay information for the group. > > > > > > That's a pain, though, since it means that userspace has to do a lot > > > more /proc scraping, and can't reliably get an accurate value for all > > > tasks, due to > > > > Delay accounting is done in kernel, summing up of data is also done in > > kernel. Collected data is sent to user space through genetlink > > interface. No /proc scraping is required. > > OK, in the paragraph above you said that userspace "sums up delay > stats for each task in the group". Can you clarify exactly how it oops.. sorry for confusing you. It is done in kernel. > works, or point me at a the code? (Which kernel version did it appear > in?) Does it automatically handle newly created or transient > processes/threads? Yes. It is in 2.6.18-rc2. You can read the documentation in Documentation/accounting/delay-accounting.txt Documentation/accounting/taskstats-struct.txt > > Paul > > ------------------------------------------------------------------------- > Using Tomcat but need to do more? Need to support web services, security? > Get stuff done quickly with pre-integrated technology to make your job easier > Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo > http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 > _______________________________________________ > ckrm-tech mailing list > https://lists.sourceforge.net/lists/listinfo/ckrm-tech -- ---------------------------------------------------------------------- Chandra Seetharaman | Be careful what you choose.... - [EMAIL PROTECTED] | .......you may get it. ---------------------------------------------------------------------- ------------------------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 _______________________________________________ ckrm-tech mailing list https://lists.sourceforge.net/lists/listinfo/ckrm-tech
