[xen-tools-discuss] Debian Wheezy as DomU see less/wrong RAM

Andy Smith andy at bitfolk.com
Tue Jan 22 21:50:33 CET 2013


Hi,

On Tue, Jan 22, 2013 at 12:27:27PM +0100, Alessio Cecchi wrote:
> I'm running Debian Wheezy as Dom0 and with xen-tool (from repository)
> I have created a new VM (Debian Wheezy) with 512MB of RAM:

[512M VM sees ~494M RAM]

AFAIK it's a difference between PVOPS and old-style Xen kernels.

There is some small overhead of RAM that is not usable by the OS,
and this also is the case on bare metal. Old style Xen kernels used
to take it from dom0, PVOPS kernels take what dom0 gives them and
deduct from what the VM gets.

Since you'll see a similar amount "missing" on bare metal, I think
the PVOPS way is more correct. Also the old way means that when you
give a VM X amount of RAM it actually uses X+Y, whereas PVOPS
kernels just use X, which is more manageable.

Example old VM running 2.6.18-6-xen-686:

Name                                        ID   Mem VCPUs      State   Time(s)
etchdemo                                   272   320     1     -b----    3237.8

$ free -m
             total       used       free     shared    buffers     cached
Mem:           320        315          4          0        119         61

PVOPS VM on same dom0, 3.2.0-2-686-pae:

Name                                        ID   Mem VCPUs      State   Time(s)
queue0                                     561   480     1     -b----  28980.2

$ free -m
             total       used       free     shared    buffers     cached
Mem:           469        391         78          0         36        212

A bare metal host:

$ free -m
             total       used       free     shared    buffers     cached
Mem:          3842       3501        340          0          0       2993

This host has 4G of RAM. Of course the more devices on a bare metal
host mean that more of the physical RAM will be reserved.

Cheers,
Andy

-- 
http://bitfolk.com/ -- No-nonsense VPS hosting


More information about the xen-tools-discuss mailing list