|
|
|
|
|
|
|
|
|
|
xen-devel
RE: [Xen-devel] linux kernel 2.6.37
> From: Vasiliy G Tolstov [mailto:v.tolstov@xxxxxxxxx]
> Sent: Tuesday, January 25, 2011 10:35 AM
> To: Konrad Rzeszutek Wilk
> Cc: Don Brearley; xen-devel@xxxxxxxxxxxxxxxxxxx
> Subject: Re: [Xen-devel] linux kernel 2.6.37
>
> On Tue, 2011-01-25 at 09:45 -0500, Konrad Rzeszutek Wilk wrote:
> > I think I know why this is happening = the
> > kernel allocates pagetables as if you have 4GB of memory and that
> eats much of low memory.
> > If you decrease the 'maxmem' you should see a smaller amount being
> used.
> >
>
> And why xenlinux kernel from centos (2.6.18-194.26.1.el5xen) displays
> all memory as available:
>
> cat /proc/meminfo under 2.6.37 (iommu=off acpi=off):
> MemTotal: 447220 kB
>
> cat /proc/meminfo under 2.6.18-194.26.1.el5xen:
> MemTotal: 524288 kB
>
> ?
FYI, I've seen a similar problem with 2.6.37 but noticed
it when manually ballooning a PV guest. For example:
# echo $((256*1024)) > /sys/devices/system/xen_memory/xen_memory0/target_kb
should reduce memory to 256M (or at least somewhere close)
but instead /proc/meminfo shows 217812 kB, or about 15%
lower than expected. This looks like approximately the
same percentage of memory loss that Vasiliy is reporting
on boot.
/me wonders if it might be related to this thread:
http://lists.xensource.com/archives/html/xen-devel/2010-09/msg00987.html
assuming the related patch got upstream. Jeremy cc'ed just in case.
I do NOT see this issue with 2.6.35.4. In fact, I now see
a MemTotal boot-time difference when I boot both with mem=512
(in vm.cfg): 2.6.35.4 kernel shows 506864, 2.6.37 shows 495456.
Not as much as Vasiliy reports, but still a big mystery.
If this problem is real and widespread, I hope nobody is comparing
KVM vs Xen (or Xen vs native) on 2.6.37 as that much memory loss
could result in a significant performance loss for a Xen guest.
Dan
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
|
|
|
|