Data Center Déjà Vu All Over Again

 Email Article  

Tweet  

  

  

  

 delicious  

  

Recently, I found myself shivering under a blast of cold air in the second aisle of a Silicon Valley data center.   I did not remember it being so cold when we visited it the month before. Two tiles to the left of where I was standing, the temperature was a tropical 100 degrees!  It turned out the aggressive server consolidation project was under way and had successfully eliminated two racks of old servers. Unfortunately, nobody told the facility manager that the air vents freezing the open area of the aisle were no longer cooling any servers.

Fast forward three weeks and it was déjà vu all over again. I was in a large data center in New England and beside boxes of new quad-core servers were new racks that were being added to the aisles and the last row of the data center was Arctic cold. Not surprisingly, this data center was also running a frenetic “rip-and-replace” server consolidation project  at 10 servers/week, and the facilities folks had not changed the vent layout yet.  Fast forward another four weeks to a data center in midtown Manhattan, and you guessed it, it was déjà vu all over again! New racks were being added to the end of two aisles and no changes had been made to the perforated tile configuration!

I cannot blame the data center managers. They are putting in place new, virtualized servers with greater compute power density. Unfortunately, they don’t know exactly how much power each rack and each server within the rack actually consumes.  So, how can they be expected to know how much cooling they need or what to tell the facility manager?   When a data center manager finds a convenient empty rack space to add new servers or he or removes racks of servers to reduce total power, he is being efficient as far as he can tell.

If the data center manager had calculated the power usage effectiveness (PUE), before my first two data center visits, he would have realized that even if the total server power decreased, the PUE may not have decreased to the same degree if the cooling efficiency increased with the more skewed power distribution.  Unfortunately, the data center manager did not get the return on investment (ROI) on the new servers.  

So, what’s the poor data center manager and the poor facilities manager to do? I believe the answer is “know their data center.”   They need to monitor the actual power consumed by the rack and by all equipment in it. We know that the data center manager will be making frequent adds, moves, and changes as business dictates. But without real-time granular power monitoring, the data center manager has no idea how much power they need and how to distribute it until, as one manager put it, “the fuse blows.”

Aloke Guha, Viridity Software

P.S. Last week, I talked to the facilities manager at the Silicon Valley data center. When he proudly mentioned how the data center was being progressive by reducing energy and deploying new computer room air conditioners (CRACs), I told him about the really cold zones. He got defensive and mentioned he did not how what kind of kW that IT was using. “What about real-time monitoring?” I asked.  “Exactly!” he said. “That’s what we need if we can get it.”