[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-users] RFC: clocksource={xen, tsc} difference between Debian PV and HV


  • To: xen-users@xxxxxxxxxxxxx
  • From: Philipp Hahn <hahn@xxxxxxxxxxxxx>
  • Date: Tue, 09 Dec 2014 08:44:23 +0100
  • Delivery-date: Tue, 09 Dec 2014 07:46:45 +0000
  • List-id: Xen user discussion <xen-users.lists.xen.org>

Hello,

On 02.12.2014 19:19, Philipp Hahn wrote:
> while running a simple "sysbench --test=memory run" benchmark I noticed
> the following strange behavior: Using "clocksource=tsc" as clock-source
> both PV and HV are nearly the same, but with "clocksource=xen" PV is 4-5
> times worse than HV. I would have expected the reverse:
...
> Is this normal and how can this be explained?

To answer my own question: There's a very good article about the
background, on why PV is slower than HVM on amd64:
<http://wiki.xen.org/wiki/Virtualization_Spectrum>: The 64 bit mode only
has 2 privilege levels, which are not enough to efficiently implement
the separation between the 3 modes required for hypervisor, kernel- and
user-mode. Because of that syscalls get a lot more expensive with PV.

Philipp

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.