Microsoft Research Paper, Measuring Energy use of a Virtual Machine

Microsoft TechFest is going on now.

About TechFest

TechFest 2010TechFest is an annual event that brings researchers from Microsoft Research’s locations around the world to Redmond to share their latest work with fellow Microsoft employees. Attendees experience some of the freshest, most innovative technologies emerging from Microsoft’s research efforts. The event provides a forum in which product teams and researchers can discuss the novel work occurring in the labs, thereby encouraging effective technology transfer into Microsoft products.

I used to go when I was a full time employee, but there are some good things you can learn by going to the demo site.

One that caught my eye is the Network Embedded Computing that has consistently worked on data center energy sensor systems.

Their latest project is Joulemeter, a project that can measure the energy usage of VM, Server, Client and Software.

Joulemeter: VM, Server, Client, and Software Energy Usage

Joulemeter is a software based mechanism to measure the energy usage of virtual machines (VMs), servers, desktops, laptops, and even individual softwares running on a computer.

Joulemeter estimates the energy usage of a VM, computer, or software by measuring the hardware resources (CPU, disk, memory, screen etc) being used and converting the resource usage to actual power usage based on automatically learned realistic power models.

Joulemeter overview

Joulemeter can be used for gaining visibility into energy use and for making several power management and provisioning decisions in data centers, client computing, and software design.

For more technical details on the system here is their paper.

Virtual Machine Power Metering and Provisioning
Aman Kansal, Feng
Zhao, Jie Liu
Microsoft Research
Nupur Kothari
University of Southern
Arka Bhattacharya
IIT Kharagpur
Virtualization is often used in cloud computing platforms for its
several advantages in efficient management of the physical resources.
However, virtualization raises certain additional challenges, and
one of them is lack of power metering for virtual machines (VMs).
Power management requirements in modern data centers have led
to most new servers providing power usage measurement in hardware
and alternate solutions exist for older servers using circuit and
outlet level measurements. However, VM power cannot be measured
purely in hardware. We present a solution for VM power metering.
We build power models to infer power consumption from resource
usage at runtime and identify the challenges that arise when
applying such models for VM power metering. We show how existing
instrumentation in server hardware and hypervisors can be
used to build the required power models on real platforms with low
error. The entire metering approach is designed to operate with
extremely low runtime overhead while providing practically useful
accuracy. We illustrate the use of the proposed metering capability
for VM power capping, leading to significant savings in power provisioning
costs that constitute a large fraction of data center power
costs. Experiments are performed on server traces from several
thousand production servers, hosting Microsoft’s real-world applications
such as Windows Live Messenger. The results show that
not only does VM power metering allows reclaiming the savings
that were earlier achieved using physical server power capping, but
also that it enables further savings in provisioning costs with virtualization.

Note there will be a desktop and laptop version available soon.

Download: A freely downloadable version of the Joulemeter software that measures laptop and desktop energy usage will be be available in a few weeks. Watch this space!

But I want to get access to the VM, Server and software versions for the data center.  Maybe I can get this group involved with GreenM3 being transition into an NPO.  The University of Missouri is also another connection as a  Mizzou professors worked with the Microsoft Researchers in a prior job developing sensor networks.