No, it’s not the Power Grid. It is the data center, and grid computing. It’s multi-processor computers running virtual machines. As one guy said to me, “Before we had a bunch of freshly delivered pizza boxes. Now we have flamethrowers in the rack”
These systems have extraordinary power requirements. These systems have extraordinary air conditioning requirements. These systems have extraordinarily bad failure modes if they overheat or lose power.
These systems are no longer managed as computers. Sure the data center uses DMTF (Desktop Management Task Force) information to track and manage them as assets.
What these systems are managed for is business services. How fast is that transaction running? Should I move the virtual computer to another system? Should I launch another virtual computer to run in parallel. Is it OK to delay processing HR to meet my QOS targets for the delivery of sales transactions?
Sales transactions are quick. What if my transactions are long running? What if the cost of losing this application is mid process is very high, and the process runs for hours? What would I do differently if I want this process to complete as quickly as possible yet the cost of failure during any single transaction is small?
These questions come down to energy. A computer working twice as hard can use much more energy than one that is not. A computer with multiple virtual machines running generates a lot of heat. Energy systems can be pushed to the brink and more by grid data centers.
If a computer has redundant power supplies, I have some choices. I can run it flat out, using all the power available from each power source and lose redundancy. I can run it half loaded and suffer failure of one power supply without noticing. Load becomes a proxy for reliability.
Of course, nothing is more efficient at converting electricity to heat than a computer. This means each of these decisions effect heat load in the building, and there by air conditioning. It take more energy to remove heat from a system than to put it in. Because air conditioning equipment can accept dirtier power than the computer can, there may be another decision about power reliability made there. Conversely, a failure of cooling systems may require shedding business services to reduce heat generation
Time of day billing, brown outs, and even black-outs require that the mix expand to include generators, batteries, and perhaps even fuel cells. If that is not enough, you can include considerations of AC vs. DC power.
These are the problems that The Green Grid is wrestling with. They wish to extend the data center’s management paradigm to include the facilities services that support the data center.
I got to sit in on one of their meetings this afternoon. Eaton was trying to formulate their technology and interface to support this vision. Aaron Merkin from IBM, well versed in DMTF, was explaining the data center side of things. I was there to talk of oBIX. Aaron is one of those guys whose knowledge of standards is deep and wide. It was a fascinating afternoon.
I wish the Green Grid the best of luck and a quick journey. I hope we can find some way for oBIX to use their surfaces to define the standard contracts for Power Systems. I know I could use their work.
Check out www.TheGreenGrid.org.