While XMission buys 100% wind power for our data center, we want to do better than simply offset our carbon footprint. We want to improve efficiencies. Not only is that always good for business, it is forward thinking.

With these concerns in mind, what is the solution? Of course, there are many things that can and should be done to help remedy these challenges. Last year, we built a return air plenum in our ceiling to better extract heat from the hot rows (we already had hot and cold rows, like most data centers). This has helped up to a point but we are next going to use blanking panels to better isolate the hot and cold rows. This not only helps get the chilled air to the computers but it minimizes bleed over into the hot rows; air conditioners can more efficiently cool 90+ degree air than 80 degree air. The Liebert systems we have in our facility can cool air down about 30 degrees but they can only make the air so cold.

We’ll see how much the blanking panels help. The next step is to fully isolate the hot and cold rows using plexiglas above the cabinets and at the end of the rows. This includes putting plexiglas doors at the end of each row, so it is a somewhat extreme measure but it may well be worth doing.

After that? Our current facility is almost full but we are evaluating expansion. If we build our own, we might go bleeding edge green, which includes using huge fans to extract the heat entirely out of the building and large evaporative coolers to help with the cooling. Mechanical engineers are seeing some success with these methods, dramatically increasing air flow to manage the heat problems. Another alternative, water cooling, is also being explored but only can work where all gear supports water cooling systems directly to the CPU. That might be a solution in the future but for now it doesn’t make sense in colocation facilities where customers have diverse systems which often can change.

Stay tuned while we adapt and look towards the future.

While there, I had the chance to hear from Adrian Cockroft, who is currently leading a performance tuning and research group over at Netflix.

Though his presentation was excellent, one of the most interesting thoughts he had was about the future of the hard drive, which he amusingly called spinning platters of rust.

Though projecting prices for storage is never a particularly exact science, some people have tried and show some interesting projections:

Given that we’ve already seem some major storage vendors moving toward solid-state arrays one has to wonder what effect this will have on operating system architecture. Cockroft pointed out that once we see effectively flat memory architectures in machines with much smaller differences in seek times between media, we may see a major effort underway to re-architect the way that operating systems see the idea of hierarchical memory.

Perhaps even more compelling from a performance standpoint is what this re-architecture might do to the idea of the database as a storage engine. For the truly curious, both Microsoft Research and Oracle Research are actively considering the question.