I sent the following mail without subscribing to the list, and I am not sure whether the list accepts mails from non-members. I just subscribed and re-sending the mail. I apologize if you receive multiple copies. Hi, I am using the credit scheduler in the xen-unstable trunk and I am trying to set a cap on the container CPU entitlement. I have two VMs and two CPUs. The two VMs are restricted to one CPU using the command xm vcpu-pin. Now, I want to cap the container CPU consumption say to 4% and 26% of the ONE CPU. So, I use the following commands. xm sched-credit -d 73 -c 4 xm sched-credit -d 74 -c 26 When I run Xentop, I see this apache.padalap.1 ------ 83 7.0 ... apache.padalap.2 -----r 154 25.5 ... How is this possible when I cap the first VM to 4% ? Am I missing some thing ? Does credit scheduler fail to enforce small caps ? Note that I haven''t changed the weights and they are set to default 256 each. Thanks, Pradeep _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Hi, Padala, The cap effects for avarage not one time. Please send xentop -b (batch) log file. Thanks Atsushi SAKAI>I sent the following mail without subscribing to the list, and I am not >sure whether the list accepts mails from non-members. I just subscribed >and re-sending the mail. I apologize if you receive multiple copies. > >Hi, > >I am using the credit scheduler in the xen-unstable trunk and I am >trying to set a cap on the container CPU entitlement. I have two VMs and >two CPUs. The two VMs are restricted to one CPU using the command xm >vcpu-pin. Now, I want to cap the container CPU consumption say to 4% and >26% of the ONE CPU. > >So, I use the following commands. > >xm sched-credit -d 73 -c 4 >xm sched-credit -d 74 -c 26 > >When I run Xentop, I see this > >apache.padalap.1 ------ 83 7.0 ... >apache.padalap.2 -----r 154 25.5 ... > >How is this possible when I cap the first VM to 4% ? Am I missing some >thing ? Does credit scheduler fail to enforce small caps ? Note that I >haven''t changed the weights and they are set to default 256 each. > >Thanks, >Pradeep > >_______________________________________________ >Xen-devel mailing list >Xen-devel@lists.xensource.com >http://lists.xensource.com/xen-devel >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> The cap effects for avarage not one time. > Please send xentop -b (batch) log file.What do you mean by average ? I am actually measuring the CPU times every 10 secs in my program. I believe Xentop averages it over 3 secs. Isn''t that long enough time to cap correctly ? A copy of xentop -b is at http://www.eecs.umich.edu/~ppadala/out. I set the caps to 4% and 96% and in the first VM, I ran a CPU eater which is basically do {;} {while(1)} I can run it for a longer period if you want. Pradeep> >I sent the following mail without subscribing to the list, > and I am not > >sure whether the list accepts mails from non-members. I just > subscribed > >and re-sending the mail. I apologize if you receive multiple copies. > > > >Hi, > > > >I am using the credit scheduler in the xen-unstable trunk and I am > >trying to set a cap on the container CPU entitlement. I have > two VMs and > >two CPUs. The two VMs are restricted to one CPU using the command xm > >vcpu-pin. Now, I want to cap the container CPU consumption > say to 4% and > >26% of the ONE CPU. > > > >So, I use the following commands. > > > >xm sched-credit -d 73 -c 4 > >xm sched-credit -d 74 -c 26 > > > >When I run Xentop, I see this > > > >apache.padalap.1 ------ 83 7.0 ... > >apache.padalap.2 -----r 154 25.5 ... > > > >How is this possible when I cap the first VM to 4% ? Am I > missing some > >thing ? Does credit scheduler fail to enforce small caps ? > Note that I > >haven''t changed the weights and they are set to default 256 each. > > > >Thanks, > >Pradeep > > > >_______________________________________________ > >Xen-devel mailing list > >Xen-devel@lists.xensource.com > >http://lists.xensource.com/xen-devel > > > > > > > > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Hi , Paradeep. Thank you for your log. and your report seems to be interesting. Please notice me the configuration of your measurement. (it seems different config to your mail.) Anyway, credit scheduler default slice is 30msec. and 1msec is the credit unit. 4% is close to 1/30. CPU intensive jobs are running 30msec at once. Under this environment, this job seems to run every 900msec(30*30). (Close to 1sec) For this environment I want to see 100sec avarage. (But currently enough by your report) Anyway thank you for your report. If you give me a configuration, I will checking on this. Thanks Atsushi SAKAI>> The cap effects for avarage not one time. >> Please send xentop -b (batch) log file. > >What do you mean by average ? I am actually measuring the CPU times >every 10 secs in my program. I believe Xentop averages it over 3 secs. >Isn''t that long enough time to cap correctly ? > >A copy of xentop -b is at http://www.eecs.umich.edu/~ppadala/out. >I set the caps to 4% and 96% and in the first VM, I ran a CPU eater >which is basically > >do {;} {while(1)} > >I can run it for a longer period if you want. > >Pradeep > >> >I sent the following mail without subscribing to the list, >> and I am not >> >sure whether the list accepts mails from non-members. I just >> subscribed >> >and re-sending the mail. I apologize if you receive multiple copies. >> > >> >Hi, >> > >> >I am using the credit scheduler in the xen-unstable trunk and I am >> >trying to set a cap on the container CPU entitlement. I have >> two VMs and >> >two CPUs. The two VMs are restricted to one CPU using the command xm >> >vcpu-pin. Now, I want to cap the container CPU consumption >> say to 4% and >> >26% of the ONE CPU. >> > >> >So, I use the following commands. >> > >> >xm sched-credit -d 73 -c 4 >> >xm sched-credit -d 74 -c 26 >> > >> >When I run Xentop, I see this >> > >> >apache.padalap.1 ------ 83 7.0 ... >> >apache.padalap.2 -----r 154 25.5 ... >> > >> >How is this possible when I cap the first VM to 4% ? Am I >> missing some >> >thing ? Does credit scheduler fail to enforce small caps ? >> Note that I >> >haven''t changed the weights and they are set to default 256 each. >> > >> >Thanks, >> >Pradeep >> > >> >_______________________________________________ >> >Xen-devel mailing list >> >Xen-devel@lists.xensource.com >> >http://lists.xensource.com/xen-devel >> > >> >> >> >> >> >> >> > >_______________________________________________ >Xen-devel mailing list >Xen-devel@lists.xensource.com >http://lists.xensource.com/xen-devel >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
What kind of configuration do you need ? CPU type, kernel etc ? Pradeep> -----Original Message----- > From: Atsushi SAKAI [mailto:sakaia@jp.fujitsu.com] > Sent: Tuesday, November 21, 2006 10:35 PM > To: Padala, Pradeep; xen-devel@lists.xensource.com > Subject: Re: [Xen-devel] Credit scheduler anamoly > > Hi , Paradeep. > > Thank you for your log. and your report seems to be interesting. > Please notice me the configuration of your measurement. > (it seems different config to your mail.) > > Anyway, credit scheduler default slice is 30msec. > and 1msec is the credit unit. > 4% is close to 1/30. > CPU intensive jobs are running 30msec at once. > Under this environment, this job seems to run every 900msec(30*30). > (Close to 1sec) > > For this environment I want to see 100sec avarage. > (But currently enough by your report) > > Anyway thank you for your report. > If you give me a configuration, I will checking on this. > > Thanks > Atsushi SAKAI > > > > > > >> The cap effects for avarage not one time. > >> Please send xentop -b (batch) log file. > > > >What do you mean by average ? I am actually measuring the CPU times > >every 10 secs in my program. I believe Xentop averages it > over 3 secs. > >Isn''t that long enough time to cap correctly ? > > > >A copy of xentop -b is at http://www.eecs.umich.edu/~ppadala/out. > >I set the caps to 4% and 96% and in the first VM, I ran a CPU eater > >which is basically > > > >do {;} {while(1)} > > > >I can run it for a longer period if you want. > > > >Pradeep > > > >> >I sent the following mail without subscribing to the list, > >> and I am not > >> >sure whether the list accepts mails from non-members. I just > >> subscribed > >> >and re-sending the mail. I apologize if you receive > multiple copies. > >> > > >> >Hi, > >> > > >> >I am using the credit scheduler in the xen-unstable trunk > and I am > >> >trying to set a cap on the container CPU entitlement. I have > >> two VMs and > >> >two CPUs. The two VMs are restricted to one CPU using the > command xm > >> >vcpu-pin. Now, I want to cap the container CPU consumption > >> say to 4% and > >> >26% of the ONE CPU. > >> > > >> >So, I use the following commands. > >> > > >> >xm sched-credit -d 73 -c 4 > >> >xm sched-credit -d 74 -c 26 > >> > > >> >When I run Xentop, I see this > >> > > >> >apache.padalap.1 ------ 83 7.0 ... > >> >apache.padalap.2 -----r 154 25.5 ... > >> > > >> >How is this possible when I cap the first VM to 4% ? Am I > >> missing some > >> >thing ? Does credit scheduler fail to enforce small caps ? > >> Note that I > >> >haven''t changed the weights and they are set to default 256 each. > >> > > >> >Thanks, > >> >Pradeep > >> > > >> >_______________________________________________ > >> >Xen-devel mailing list > >> >Xen-devel@lists.xensource.com > >> >http://lists.xensource.com/xen-devel > >> > > >> > >> > >> > >> > >> > >> > >> > > > >_______________________________________________ > >Xen-devel mailing list > >Xen-devel@lists.xensource.com > >http://lists.xensource.com/xen-devel > > > > > > > > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Are you running a xen-unstable build containing this changeset from November 1st? There was a bug in the cap mechanism -- found by your colleagues at HP -- when a physical CPU was contended. This changeset seems to have made things better. changeset: 12202:9a4274724794 user: Emmanuel Ackaouy <ack@xensource.com> date: Wed Nov 01 19:44:34 2006 +0000 summary: [XEN] Fix credit scheduler cap mechanism not to over park VCPUs On Tue, Nov 21, 2006 at 04:07:00PM -0600, Padala, Pradeep wrote:> I sent the following mail without subscribing to the list, and I am not > sure whether the list accepts mails from non-members. I just subscribed > and re-sending the mail. I apologize if you receive multiple copies. > > Hi, > > I am using the credit scheduler in the xen-unstable trunk and I am > trying to set a cap on the container CPU entitlement. I have two VMs and > two CPUs. The two VMs are restricted to one CPU using the command xm > vcpu-pin. Now, I want to cap the container CPU consumption say to 4% and > 26% of the ONE CPU. > > So, I use the following commands. > > xm sched-credit -d 73 -c 4 > xm sched-credit -d 74 -c 26 > > When I run Xentop, I see this > > apache.padalap.1 ------ 83 7.0 ... > apache.padalap.2 -----r 154 25.5 ... > > How is this possible when I cap the first VM to 4% ? Am I missing some > thing ? Does credit scheduler fail to enforce small caps ? Note that I > haven''t changed the weights and they are set to default 256 each. > > Thanks, > Pradeep > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
I will get the latest xen-unstable and try it again. Thanks ! Pradeep> -----Original Message----- > From: Emmanuel Ackaouy [mailto:ack@xensource.com] > Sent: Wednesday, November 22, 2006 4:57 AM > To: Padala, Pradeep > Cc: xen-devel@lists.xensource.com > Subject: Re: [Xen-devel] Credit scheduler anamoly > > Are you running a xen-unstable build containing this > changeset from November 1st? > > There was a bug in the cap mechanism -- found by your > colleagues at HP -- when a physical CPU was contended. This > changeset seems to have made things better. > > changeset: 12202:9a4274724794 > user: Emmanuel Ackaouy <ack@xensource.com> > date: Wed Nov 01 19:44:34 2006 +0000 > summary: [XEN] Fix credit scheduler cap mechanism not to > over park VCPUs > > > On Tue, Nov 21, 2006 at 04:07:00PM -0600, Padala, Pradeep wrote: > > I sent the following mail without subscribing to the list, and I am > > not sure whether the list accepts mails from non-members. I just > > subscribed and re-sending the mail. I apologize if you > receive multiple copies. > > > > Hi, > > > > I am using the credit scheduler in the xen-unstable trunk and I am > > trying to set a cap on the container CPU entitlement. I > have two VMs > > and two CPUs. The two VMs are restricted to one CPU using > the command > > xm vcpu-pin. Now, I want to cap the container CPU > consumption say to > > 4% and 26% of the ONE CPU. > > > > So, I use the following commands. > > > > xm sched-credit -d 73 -c 4 > > xm sched-credit -d 74 -c 26 > > > > When I run Xentop, I see this > > > > apache.padalap.1 ------ 83 7.0 ... > > apache.padalap.2 -----r 154 25.5 ... > > > > How is this possible when I cap the first VM to 4% ? Am I > missing some > > thing ? Does credit scheduler fail to enforce small caps ? > Note that I > > haven''t changed the weights and they are set to default 256 each. > > > > Thanks, > > Pradeep > > > > _______________________________________________ > > Xen-devel mailing list > > Xen-devel@lists.xensource.com > > http://lists.xensource.com/xen-devel >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel