[gridvirt-wg] Protocol: Open discussion on GridVirt-WG in OGF23

[Bozak, Erol]

FYI...

________________________________

From: Bozak, Erol 
Sent: Mittwoch, 16. Juli 2008 13:51
To: 'lee at aero.org'; ftravostino at ebay.com; Hughes-Jones at man.ac.uk;
csmith at platform.com; werstiuk at platform.com; stokese at us.ibm.com
Cc: Reichert, Wolfgang (external); Wusthoff, Mark;
craig at mhbackup.aero.org
Subject: RE: Protocol: Open discussion on GridVirt-WG in OGF23


Hello Lee,
 
sounds interesting. The agenda for today is to agree on the work
packages on which we will focus on in the specification document. 
So I think this will consume most of the time in the session.
 
I propose that we have a follow-up on this in the next call discussing
your proposal explicitly.
 
Let me give you a quick answer to your proposal:
One requirement of the interface(s) the working group is looking into is
dealing with policies (such as proposed below). In the course of this
work package I think we can explore this topic. However, I do not think
it is the main scope of this working group to define / specify the
architecture & algorithm to enforce this policy. Additionally, the
interaction among a virtualization manager and the sensors to measure
power consumption metrics is, I think, also not scope. Basically, policy
enforcement is a platform specific detail that is hidden by the
management layer and thus not explicitly exposed. 
 
However, we can have a detailed discussion in a dedicated concall.
 
Regards, Erol

________________________________

From: lee at aero.org [mailto:lee at aero.org] 
Sent: Dienstag, 15. Juli 2008 16:51
To: Bozak, Erol; ftravostino at ebay.com; Hughes-Jones at man.ac.uk;
csmith at platform.com; werstiuk at platform.com; stokese at us.ibm.com
Cc: Reichert, Wolfgang (external); craig at mhbackup.aero.org
Subject: Re: Protocol: Open discussion on GridVirt-WG in OGF23
Importance: High



Erol, et al.,

I will be attending tomorrow's telecon and would like to see if there
is any interest in pursuing a project along the lines of what's proposed
below.  This is part of an effort to use "targeted projects" as a way
to produce "value" for members.  I've attached a short slide deck
that describes the overall process and how it might work.  It also
occurs to me that perhaps we can get the RESERVOIR project
interested.  Kate Keahey from Argonne is already interested.

Let's talk tomorrow, if we can fit this into the agenda...

Best regards,

--Craig

==================================================

Project Proposal:  Using Virtual Machines to Enact Energy Policy

Goal:

The goal of this proposed project is to study and demonstrate the
realization and enforcement of energy policy, i.e., Green IT, through
the initial placement and migration of virtual machines (VMs).

Why it it important:

The management of energy utilization has rapidly become a
critical issue for industry, government, and academia, since
many computing installations are exceeding the available
electrical supply, cooling and floor space.  The use of VMs
to dynamically manage workload location and distribution
is considered to be a highly promising method for managing
energy efficiency and meeting corporate requirements.

To Whom This is Important:

This project would benefit all members that are concerned about
maximizing the cost-effectiveness and efficiency of data center
operation.  Since this approach is inherently distributed, it may
incorporate and rely upon many traditional grid computing concepts.

Technical Description:

Managing energy efficiency to achieve an energy policy would be a
matter of:

(1) Integration of appropriate monitoring capabilities to monitor
    current energy consumption at a given site, e.g., cooling,
    processors, storage, etc.,

(2) The definition of semantically complete and consistent
    representations of energy policy, possibly using XML,

(3) The use of (a) policy representations and (b) current (monitored)
    system state, to direct (c) VM placement and migration to achieve
    and enforce energy policy.  This may be done through a set of
    energy management peers that monitor the energy usage at their own
    site, gossip among themselves about how workload should be
    distributed, and act accordingly.

This project could possibly leverage capabilities already developed by
personnel at Argonne National Laboratory (ANL).  ANL has worked on the
concept of virtual workspaces and virtual clusters where sets of VMs
are managed together and yet may be distributed across remote physical
resources.  This is very attractive since exploiting existing
capabilities
always makes it easier to quickly demonstrate new capabilities.

Required Resources:

This project would require at least two sites where VMs could be
dynamically managed.  More sites would be preferable, but not to the
point where it becomes unmanageable for a targeted demonstration.  A
user workload would have to be generated, either synthetically or
through the use of real applications, that would provide VMs that need
to be allocated.  Personnel would have to be available and supported to:

1) Build and integrate (or possibly simulate) the monitoring subsystem,

2) Devise an energy policy representation,

3) Build the p2p control subsystem that would use the monitored
   information and the energy policy to determine VM initial placement
   and migration, and

4) perform experiments to verify the effectiveness of this approach of
   enforcing energy policy.

Time Frame:

It is expected that with the right provisioning, this project could be
completed within six months.  Each of the subtask identified above
could be done is the simplest manner possible to demonstrate a
proof of concept implementation that leverages existing capabilities
for VM placement, etc.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.ogf.org/pipermail/gridvirt-wg/attachments/20080716/8349dd75/attachment.html