[CentOS-virt] Pages In and Read Sectors in Guest Domain

Tue Apr 29 15:35:16 UTC 2014
Deron <fecastle at gmail.com>

I think I know why this is happening, and I think it is correct.   I ran a
test again only looking at Dom0 and found that it was similar to Dom1.  The
'read sectors' from /proc/diskstats were about 2x the pgpgin from
/proc/vmstat.   This is whay I see on my physical hardware as well.

The reason that I am not seeing the same results when reading data from
Dom1 is because I am only looking at the physical block device (/dev/sda in
this case).    When I look at the diskstats from physical device /dev/sda
and device between used for blkback /dev/tda they show the 2 reads.  Since
they are both "block devices" I assume they need to page in to make the
data available.

So I think this is not really an issue, just a matter of perspective.

Deron



On Wed, Apr 9, 2014 at 4:15 AM, Stefano Stabellini <
stefano.stabellini at eu.citrix.com> wrote:

> On Sun, 6 Apr 2014, Deron wrote:
> > Hello.
> >
> > I have been doing some performance tests with Xen and CentOS and have
> found some strange statistics when comparing the
> > virtual memory (/proc/vmstat) and I/O (/proc/diskstat) statistics
> between the Domain 0 and guests.
> >
> > When I look at the "pgpgin" statistic it is exactly the same as the
> "read sectors" in Dom0.  However, when I look at
> > the same statistics in the guest, the "read sectors" is exactly 2x
> "pgpgin".
> >
> > I have tried this on 2 different versions, each on a different physical
> server. Here is an example of the statistics
> > over a 30 second period (while running a benchmark in Dom1) in the Dom0
> and Guest Domain.
> >                        pgpgin       read sectors
> >                        =======     =========
> >       Dom0        1698024     1698024
> >       Dom1          848412     1696824
> >
> > I have looked at some various settings to see if I could explain it.  I
> can't tell if the Dom1 guest kernel is
> > reporting incorrectly, or there is some other explanation.  If anyone
> has any ideas about this I would appreciate any
> > feedback.
> >
> > Dom 0 kernels
> > 3.10.29-11.el6.centos.alt.x86_64
> > 3.4.54-8.el6.centos.alt.x86_64
> >
> > Guest kernel
> > 2.6.32-431.el6.x86_64
>
> The numbers that you are seeing are probably wrong due to an accounting
> issue in blkfront or blkback. It would be interesting to know whether
> this happens even with the most recent kernels.
> _______________________________________________
> CentOS-virt mailing list
> CentOS-virt at centos.org
> http://lists.centos.org/mailman/listinfo/centos-virt
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.centos.org/pipermail/centos-virt/attachments/20140429/2133eeed/attachment-0006.html>