[114529] in North American Network Operators' Group
RE: questions about DVFS in saving energy
daemon@ATHENA.MIT.EDU (Tomas L. Byrnes)
Thu May 14 12:59:00 2009
Date: Thu, 14 May 2009 09:19:59 -0700
In-Reply-To: <4A0BD1D5.60600@theangryangel.co.uk>
From: "Tomas L. Byrnes" <tomb@byrneit.net>
To: "Karl Southern" <karl@theangryangel.co.uk>
Cc: nanog@merit.edu
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
Apologies for skirting close, but I think power consumption and heat
dissipation are pretty big operator costs, and anything we can do to
reduce those are beneficial to the bottom line; never mind the
environment. More below:
>-----Original Message-----
>From: Karl Southern [mailto:karl@theangryangel.co.uk]
>Sent: Thursday, May 14, 2009 1:10 AM
>To: Tomas L. Byrnes
>Cc: nanog@merit.edu
>Subject: Re: questions about DVFS in saving energy
>
>Tomas L. Byrnes wrote:
>> Basically the CPU scaling on the host makes the guest OS fall apart.
>>
>Apologies for the general noise (and even more apologies for stepping
>outside of the nanog scope), but if it's timing related issues does
>/usepmtimer not resolve this issue for the VMs? It certainly does on
>other virtualisation solutions.
[TLB:] We tried all the solutions we could Google, including
/usepmtimer. A potential 50% reduction in power per system (which is
what we were measuring in the tests) would be significant.
Unfortunately, it was not stable. It appears to be a Win2K3 issue,
although Red Hat Enterprise ran at the declock speed all the time, even
under heavy loads (it didn't crash and corrupt volumes like Win2K3,
however).
