HPCwire has a relatively new feature called “Community Voices,” which I kind of like — they are picking up a few folks who are already blogging in the community and aggregating the best of their stuff on the HPCwire blog site. A post in that section caught my eye this morning; it’s from Mark Thiele, the Director Business Operations at VMware.
In five years the vast majority of datacenters around today will seem like dinosaurs. You will be able to call up new capacity in 4-6 weeks and you’ll manage your environment to an efficiency level only achievable through the removal of dedicated staffing. The question is, how do you get there and how long will it take you? Your datacenter ownership strategy will be one of the critical keys to your success.
What’s Marks’ vision for the future?
- A distributed, but highly efficient datacenter footprint that is dark most of the time.
- Compute infrastructure that is portable and whose individual components make up capacity, not capability, which allows for fail in place and replacement on a schedule.
- Low cost of ownership through efficient space acquisition (you should no longer be buying datacenter space that costs you money for years before being put into use).
- Removal of forced air cooling, including potentially replacing server fans with liquid cooling.
- Greatly improved flexibility for building what you need, in the quantity you need, when you need it.
I don’t think it’s a given that all of these will be present in supercomputing datacenters, but they might be. I usually advise people to think carefully about a new technology before they declare it to be useless for anyone that “really cares” about HPC. That “really cares” part is often code for “I’ve given this exactly 12 seconds of thought, and from the paragraph I read it’s obvious this doesn’t apply to us.”