The Open Source Grid and Cluster Conference is being held
this week in Oakland, California. I attended the first day of the conference
before flying home to meet a personal commitment. My favorite talk of the day was
Paul Brenner's presentation titled Grid Heating: Dynamic Thermal
Allocation via Grid Engine Tools.
Brenner, who works as a scientist in the University of Notre Dame's
Center for Research Computing, is exploring
innovative ways to exploit the waste heat generated by HPC and
other datacenters via partnerships with various municipal entities
in the South Bend area. His first prototype, currently
in progress, involves placing a rack of HPC compute nodes at a local
municipal greenhouse, the South Bend Greenhouse and Botanical
The greenhouse had recently been forced to close portion of its facility
due to high natural gas heating costs. Brenner wondered if he could help.
Since current datacenters can be viewed as massive electricity-to-heat
converters (with a computational byproduct), it seemed there might be
an opportunity to exploit the waste heat in some useful way. But transferring
heat, especially low-grade waste heat, over distances is very inefficient.
Was there a way to overcome this barrier?
Enter grid computing with its ability to harness remotely located compute
resources. If Brenner couldn't transport the heat to the greenhouse, why
not place the datacenter at the greenhouse? The garden gets the heat and
Notre Dame gets the compute resources via established grid computing
capabilities like Sun's Grid Engine
distributed resource manager, which is already in use at Notre Dame. Cool idea? Hot idea!
Based on early prototype work which involves placing single rack in
the greenhouse, the idea looks like a promising way to reduce natural
gas heating requirements for the facility. Brenner has shown he can
use grid scheduling software to deliver a desired temperature (within
a range, of course) by simply adding or throttling compute jobs on
the greenhouse cluster, which communicates with Notre Dame via
a wide-area wireless broadband connection.
He has looked at humidity issues and so far they don't seem to be a
problem given the ranges supported by typical compute gear. And
he points out that while the greenhouse environment does not offer
the highly filtered environment of a controlled datacenter, the
particulate tolerance for typical compute gear
is far in excess of EPA guidelines for people.
Phase II will involve placing three full racks of gear at the greenhouse
to significantly reduce heating costs. Notre Dame will pay the electrical
costs and use the compute resources. The city saves money on heating.
While the greenhouse is an interesting experiment, it is not ideal since its
heating requirements will fluctuate seasonally. There are, however, other
installations that have constant heating requirements--for example,
hospitals have a 24x7 need for hot water. Sites like this could be
interesting for future deployments.
Brenner's full presentation is available [PDF].