Subject: Re: [vserver] vserver guest memory usage: cache/buffers?
From: Steve Kieu <msh.computing@gmail.com>
Date: Tue, 29 Nov 2011 10:58:38 +1100
Tue, 29 Nov 2011 10:58:38 +1100
what is the error - steps that you could not use cgroup?

I use the patch for 2.6.32.38 kernel (not really new) but it works out of
the box. The vserver tool 0.30.216.pre2986 compiled form source. I still
have the rpm here if you need to test.

cheers

On Tue, Nov 29, 2011 at 5:13 AM, Herbert Poetzl <herbert@13thfloor.at>wrote:

> On Mon, Nov 28, 2011 at 05:39:44PM +0100, Jarry wrote:
> > Hi,
>
> > finally I managed memory limits to work using the old way
> > using rss.hard and rss.soft (for whatever reason, cgroups
> > simply did not work for me). But I'm a little surprised how
> > memory is used/reported in this version (2.3.0.36.32):
>
> from the version you list, I conclude that you are using
> linux 2.6.35.2 - 2.6.35.7.
>
> the Linux-VServer memory accounting was dropped in favor
> of cgroups and 2.6.34.4-vs2.3.0.36.30.4 added a fake
> entry to keep ancient tools happy (config option)
>
> config VSERVER_LEGACY_MEM
>       bool    "Legacy Memory Limits"
>       default n
>       help
>         This provides fake memory limits to keep
>         older tools happy in the face of memory
>         cgroups
>
> > vs6a / # free
> >         total      used    free    shared  buffers  cached
> > Mem:    524288     160324  363964  0       0        158424
> > -/+ buffers/cache: 1900 522388
> > Swap:   524288     0       524288
>
> > I'm running one more older vserver-host (2.2.0.7), and there
> > I never get anything for "cached":
>
> > vs6b / # free
> >         total       used   free    shared  buffers  cached
> > Mem:    262144      6556   255588  0       0        0
> > -/+ buffers/cache:  6556   255588
> > Swap:   786432      0      786432
>
> > I also see RSS-memory usage as reported by vserver-stat is
> > now much higher:
>
> mostly because if you do not use cgroups, you get
> no accurate memory accounting at all, and util-vserver
> falls back to simply summing up the guest processes
>
> > (new) vserver-stat:
> > CTX   PROC    VSZ    RSS  userTIME   sysTIME    UPTIME NAME
> > 6000     3  88.5M 155.9M   0m15s54   0m01s94   8m11s53 vs6a
>
> > (old) vserver-stat:
> > CTX   PROC    VSZ    RSS  userTIME   sysTIME    UPTIME NAME
> > 6000     3  90.7M     4M   0m01s57   0m00s65   1d03h06 vs6b
>
> > There is absolutely the same software running (namely nothing
> > but init, cron and syslog-ng), yet the old vserver-guest takes
> > just 4MB of RSS, the new one takes 156MB! This corresponds
> > with ~150MB used for disk-cache.
>
> > Unfortunatelly, I can't compare memory usage of vserver-hosts,
> > as there is different software running. So my question is:
> > Was something changed in caching, concerning vserver-guests
> > recently? Was it "moved" from vserver-host to vserver-guests?
>
> no, nothing changed in caching, but if you want guest
> specific memory accounting, you absolutely need to
> enable and configure cgroups ... I'd also advise to
> update to a more recent kernel/patch ...
>
> best,
> Herbert
>
> > Jarry
>
> > --
> > _______________________________________________________________
> > This mailbox accepts e-mails only from selected mailing-lists!
> > Everything else is considered to be spam and therefore deleted.
>



-- 
Steve Kieu


what is the error - steps that you could not use cgroup?

I use the patch for 2.6.32.38 kernel (not really new) but it works out of the box. The vserver tool 0.30.216.pre2986 compiled form source. I still have the rpm here if you need to test.

cheers

On Tue, Nov 29, 2011 at 5:13 AM, Herbert Poetzl <herbert@13thfloor.at> wrote:
On Mon, Nov 28, 2011 at 05:39:44PM +0100, Jarry wrote:
> Hi,

> finally I managed memory limits to work using the old way
> using rss.hard and rss.soft (for whatever reason, cgroups
> simply did not work for me). But I'm a little surprised how
> memory is used/reported in this version (2.3.0.36.32):

from the version you list, I conclude that you are using
linux 2.6.35.2 - 2.6.35.7.

the Linux-VServer memory accounting was dropped in favor
of cgroups and 2.6.34.4-vs2.3.0.36.30.4 added a fake
entry to keep ancient tools happy (config option)

config VSERVER LEGACY MEM
      bool    "Legacy Memory Limits"
      default n
      help
        This provides fake memory limits to keep
        older tools happy in the face of memory
        cgroups

> vs6a / # free
>         total      used    free    shared  buffers  cached
> Mem:    524288     160324  363964  0       0        158424
> -/+ buffers/cache: 1900 522388
> Swap:   524288     0       524288

> I'm running one more older vserver-host (2.2.0.7), and there
> I never get anything for "cached":

> vs6b / # free
>         total       used   free    shared  buffers  cached
> Mem:    262144      6556   255588  0       0        0
> -/+ buffers/cache:  6556   255588
> Swap:   786432      0      786432

> I also see RSS-memory usage as reported by vserver-stat is
> now much higher:

mostly because if you do not use cgroups, you get
no accurate memory accounting at all, and util-vserver
falls back to simply summing up the guest processes

> (new) vserver-stat:
> CTX   PROC    VSZ    RSS  userTIME   sysTIME    UPTIME NAME
> 6000     3  88.5M 155.9M   0m15s54   0m01s94   8m11s53 vs6a

> (old) vserver-stat:
> CTX   PROC    VSZ    RSS  userTIME   sysTIME    UPTIME NAME
> 6000     3  90.7M     4M   0m01s57   0m00s65   1d03h06 vs6b

> There is absolutely the same software running (namely nothing
> but init, cron and syslog-ng), yet the old vserver-guest takes
> just 4MB of RSS, the new one takes 156MB! This corresponds
> with ~150MB used for disk-cache.

> Unfortunatelly, I can't compare memory usage of vserver-hosts,
> as there is different software running. So my question is:
> Was something changed in caching, concerning vserver-guests
> recently? Was it "moved" from vserver-host to vserver-guests?

no, nothing changed in caching, but if you want guest
specific memory accounting, you absolutely need to
enable and configure cgroups ... I'd also advise to
update to a more recent kernel/patch ...

best,
Herbert

> Jarry

> --
>
> This mailbox accepts e-mails only from selected mailing-lists!
> Everything else is considered to be spam and therefore deleted.



--
Steve Kieu