Re: [Vserver] Vserver CPU limit question

From: Herbert Poetzl <herbert_at_13thfloor.at>
Date: Mon 19 Mar 2007 - 23:57:22 GMT
Message-ID: <20070319235722.GD3389@MAIL.13thfloor.at>

On Mon, Mar 19, 2007 at 11:40:01PM +0100, Herbert Poetzl wrote:
> On Mon, Mar 19, 2007 at 01:52:42PM -0700, Albert Mak (almak) wrote:
> > Hi Herbert,
> >
> > I repeated the same expriment with sched_hard. The result is the
> > same, vserver is not able to enforce the CPU limit. I am under the
> > impression that sched_prio will also make use of the priority scheme
> > to limit CPU utilization per Vserver context....
>
> sounds really strange, as it is working fine here ...
> (with linux-2.6.19.7-vs2.2.0-rc19)
>
> here is a short example how you can test it, eliminating
> all possible reasons for doing something wrong
>
> - get and compile the vcmd tool [1] and the cpuhog [2]
> - do the following incantations:
>
> vcmd -i 100 -BC ctx_create .flagword=^34^33^32^8 -- cpuhog
>
> - check the results with 'vtop' which should show something
> like this:
>
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> 29 root 25 0 1312 252 200 H 24.5 0.5 0:33.81 cpuhog
> 30 root 16 0 1808 900 728 R 1.5 1.6 0:14.84 top
>
> by default, the CPU limit will be roughly 25% without
> doing any adjustments to the token buckets ...
>
> also note that a working token bucket looks like this:
>
> FillRate: 1,1
> Interval: 4,8
> TokensMin: 6
> TokensMax: 50
> PrioBias: 0
> cpu 0: 5296 11 17101 5288 0 R- 6 6 50 1/4 1/8 0 0
> ~~~~~~ hold ticks
>
> I will check that with your ancient kernel and patch
> version shortly ...

tested now with 2.6.14.3-vs2.0.1 ...

works fine here as expected:

  PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
   22 root 25 0 1304 252 200 R 24.1 0.4 0:09.59 cpuhog
   23 root 17 0 1800 896 728 R 2.5 1.5 0:02.72 top

best,
Herbert

> HTH,
> Herbert
>
> [1] http://vserver.13thfloor.at/Experimental/TOOLS/vcmd-0.08.tar.bz2
> [2] http://vserver.13thfloor.at/Experimental/TOOLS/cpuhog.c
>
> > Thanks for your help.
> > -Albert
> >
> > -bash-2.05b# cat /proc/virtual/2/status
> > UseCnt: 9
> > Tasks: 3
> > Flags: 0000000202020110
> > BCaps: 00000000354c24ff
> > CCaps: 0000000000000101
> > Ticks: 0
> >
> > -bash-2.05b# cat /proc/virtual/3/status
> > UseCnt: 9
> > Tasks: 3
> > Flags: 0000000202020110
> > BCaps: 00000000354c24ff
> > CCaps: 0000000000000101
> > Ticks: 0
> >
> >
> > top - 14:02:25 up 2:34, 3 users, load average: 1.91, 0.88, 0.34
> > Tasks: 132 total, 3 running, 129 sleeping, 0 stopped, 0 zombie
> > Cpu(s): 100.0% us, 0.0% sy, 0.0% ni, 0.0% id, 0.0% wa, 0.0% hi, 0.0% si
> > Mem: 513084k total, 118572k used, 394512k free, 16704k buffers
> > Swap: 0k total, 0k used, 0k free, 46648k cached
> >
> > PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> > 32600 root 25 0 1336 232 184 R 49.8 0.0 1:38.31 exceed_cpu_limi
> > 32697 root 25 0 1336 232 184 R 49.8 0.0 1:14.18 exceed_cpu_limi
> >
> >
> > -bash-2.05b# cat /proc/virtual/2/sched
> > Token: 140
> > FillRate: 80
> > Interval: 100
> > TokensMin: 50
> > TokensMax: 140
> > PrioBias: 0
> > VaVaVoom: 0
> > cpu 0: 127657 47 0
> >
> > -bash-2.05b# cat /proc/virtual/3/sched
> > Token: 140
> > FillRate: 10
> > Interval: 100
> > TokensMin: 50
> > TokensMax: 140
> > PrioBias: 0
> > VaVaVoom: 0
> > cpu 0: 113825 45 0
> >
> >
> >
> > -----Original Message-----
> > From: Herbert Poetzl [mailto:herbert@13thfloor.at]
> > Sent: Sun 3/18/2007 7:45 AM
> > To: Albert Mak (almak)
> > Cc: vserver@list.linux-vserver.org
> > Subject: Re: [Vserver] Vserver CPU limit question
> >
> > On Sat, Mar 17, 2007 at 10:17:47PM -0700, Albert Mak (almak) wrote:
> > > Hi Herbert
> > >
> > > Here is the output of /proc/virtual/2/status as requested.... Both
> > > context 2 and 3 have the same setting.
> > >
> > > -bash-2.05b# cat /proc/virtual/2/status
> > > UseCnt: 7
> > > Tasks: 2
> > > Flags: 0000000202020210
> > ~~~~~~~~~~
> > http://linux-vserver.org/Capabilities_and_Flags
> >
> > 0000000000000100 sched_hard
> > 0000000000000200 sched_prio
> >
> > so you haven't enabled sched_hard here, which explains
> > why you do not see hard scheduling behaviour :)
> >
> > HTC,
> > Herbert
> >
> > > BCaps: 00000000354c24ff
> > > CCaps: 0000000000000101
> > > Ticks: 0
> > >
> > > Thanks.
> > >
> > > -Albert
> > > -----Original Message-----
> > > From: Herbert Poetzl [mailto:herbert@13thfloor.at]
> > > Sent: Saturday, March 17, 2007 11:36 AM
> > > To: Albert Mak (almak)
> > > Cc: vserver@list.linux-vserver.org
> > > Subject: Re: [Vserver] Vserver CPU limit question
> > >
> > > On Fri, Mar 16, 2007 at 06:54:26PM -0700, Albert Mak (almak) wrote:
> > > > Hi,
> > > >
> > > > I have Linux (2.6.14.3 Kernel) with Vserver 2.0.1 and testing the CPU
> > > > limit capabilities. I have 2 vserver contexts both running CPU
> > > > intensive app capable of using up 100% CPU, I am setting up on vserver
> > >
> > > > to limit 1 context to 10% CPU and the 2nd to 80% CPU, both using
> > > flags sched_prio.
> > > > I am seeing CPU usage split 50/50 between the 2 contexts. I repeated
> > > > the same test using sched_hard with the same result (kernel
> > > > VSERVER_HARDCPU config set to y). I am expecting to see at least the
> > > > CPU usage close to the Vserver limits.
> > > >
> > > > Have I got the wrong settings or some other issues. Your help is
> > > > really appreciated.
> > > >
> > > > -Albert
> > > >
> > > > top - 18:37:04 up 26 min, 1 user, load average: 2.04, 1.40, 0.62
> > > > Tasks: 127 total, 3 running, 124 sleeping, 0 stopped, 0 zombie
> > > > Cpu(s): 98.7% us, 1.3% sy, 0.0% ni, 0.0% id, 0.0% wa, 0.0% hi,
> > > > 0.0% si
> > > > Mem: 513084k total, 115660k used, 397424k free, 10200k
> > > buffers
> > > > Swap: 0k total, 0k used, 0k free, 39332k
> > > cached
> > > >
> > > > PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> > > > 6616 root 20 0 1332 228 184 R 49.8 0.0 2:23.12
> > > > exceed_cpu_limi
> > > > 6513 root 20 0 1336 232 184 R 48.1 0.0 2:43.79
> > > > exceed_cpu_limi
> > > >
> > > > -bash-2.05b# vps
> > > > PID CONTEXT TTY TIME CMD
> > > > 3672 0 MAIN pts/0 00:00:00 bash
> > > > 6513 2 APP1 pts/0 00:03:01 exceed_cpu_limi
> > > > 6616 3 APP2 pts/0 00:02:40 exceed_cpu_limi
> > > > 7655 1 ALL_PROC pts/0 00:00:00 vps
> > > > 7656 1 ALL_PROC pts/0 00:00:00 ps
> > > >
> > > > -bash-2.05b# pwd
> > > > /etc/vservers/APP1
> > > > -bash-2.05b# cat flags
> > > > sched_prio
> > >
> > > you want to add sched_hard here if you want hard scheduling, the prio
> > > scheduler will only adjust priorities according to the token buckets ...
> > >
> > > I'd also suggest to use a more recent kernel (and probably Linux-VServer
> > > patch) than this one as the scheduler was enhanced quite a lot in 2.2.x
> > >
> > > > -bash-2.05b# cat schedule
> > > > 80
> > > > 100
> > > > 200
> > > > 50
> > > > 140
> > > > dummy
> > > >
> > > > -bash-2.05b# pwd
> > > > /etc/vservers/APP2
> > > > -bash-2.05b# cat flags
> > > > sched_prio
> > > > -bash-2.05b# cat schedule
> > > > 10
> > > > 100
> > > > 200
> > > > 50
> > > > 140
> > > > dummy
> > > >
> > > > -bash-2.05b# cat /proc/virtual/2/sched
> > > > Token: 140
> > > > FillRate: 1
> > > > Interval: 100
> > > > TokensMin: 50
> > > > TokensMax: 140
> > > > PrioBias: 0
> > > > VaVaVoom: -5
> > > > cpu 0: 229674 71 0
> > > >
> > > > -bash-2.05b# cat /proc/virtual/3/sched
> > > > Token: 140
> > > > FillRate: 10
> > > > Interval: 100
> > > > TokensMin: 50
> > > > TokensMax: 140
> > > > PrioBias: 0
> > > > VaVaVoom: -5
> > > > cpu 0: 217275 54 0
> > >
> > > looks like none of the token buckets is active here, what does the
> > > /proc/virtual/2/status show?
> > >
> > > TIA,
> > > Herbert
> > >
> > > > _______________________________________________
> > > > Vserver mailing list
> > > > Vserver@list.linux-vserver.org
> > > > http://list.linux-vserver.org/mailman/listinfo/vserver
> >
> _______________________________________________
> Vserver mailing list
> Vserver@list.linux-vserver.org
> http://list.linux-vserver.org/mailman/listinfo/vserver
_______________________________________________
Vserver mailing list
Vserver@list.linux-vserver.org
http://list.linux-vserver.org/mailman/listinfo/vserver
Received on Tue Mar 20 01:17:38 2007

[Next/Previous Months] [Main vserver Project Homepage] [Howto Subscribe/Unsubscribe] [Paul Sladen's vserver stuff]
Generated on Tue 20 Mar 2007 - 01:17:43 GMT by hypermail 2.1.8