How Cold Should My Cold Aisles Be?

I was recently asked a question about how cool the cold aisle should be in a data center using hot/cold aisles to manage temperature.

This question is difficult to answer without other background information.  I’ve seen specifications reference ASHRAE guidelines, Uptime Institute recommendations, and so on.  For me, TIA-942 (which is the basis for the international standard on data center design), is a good framework to lean upon.

TIA-942, Appendix G states the following for the capabilities of the raised floor area environmental air:

• 20 degrees C to 25 degrees C (68F to 77F)
  • Normal set point: 22 degrees C (72F)  +/- 1 degree C (2F)
• Relative Humidity: 40% to 55%
  • Normal set point: 45% RH +/- 5%

Now, all that may be easier said than done, of course.  The whole point of the cooling is to keep the equipment happy (operating in the prescribed temperature range).  Heat is generated by the equipment’s consumption of electricity, the level of utilization, and the density of the technology used in your data center (which may also vary by location on the raised floor, creating a non-uniform temperature pattern). 

This is also modulated by the power density capabilities of the data center’s MEP infrastructure.  If you have a relatively old facility, chances are that the power density capabilities of the facility are less than 50 Watts per Square Foot (W/SF).  If you have a modern facility, you may enjoy capacities of 85 W/SF, 100 W/SF, or even 150 W/SF (higher than that are somewhat rare these days).  A well designed MEP infrastructure has cooling capacity matched to power density capacity. 

Regardless of that though, it’s possible that you may use technologies that squeeze a very high amount of power consumption (and heat generation) into a small area.  Hot/Cold aisle arrangements may not be enough to cool those spots, and may cause consideration of supplemental cooling on the raised floor.  There are several in-row cooling technologies with which you can place a floor-standing cooling unit in-row with your equipment racks to deliver extra cooling capacity to selected areas.  One should be careful when using supplemental approaches though, that the redundancy of MEP infrastructure supporting your intended Tier Level is not obviated by such an implementation.

Do you have any experiences with specifying hot/cold aisle metrics that you’d like to share?

 

 

New Data Center Tier Classification Guidelines from the Uptime Institute

The Uptime Institute recently released an update to its white paper defining the Tier levels for data centers.   If you’re interested in looking at it, it can be found here.  This update from the Institute is a welcomed release indeed. 

In our recent experience with Clients planning, renovating, and building data centers, the earlier release of this document (over one year ago) grew to be a reference of “requirements,” “ standards,” “ best practices,” and “mandates” for leaders charged with charting the IT facility strategy and selling/supporting the ROI basis for that strategy.  The reason I put the terms above in quotes is that a fair amount of abstraction and indirect dot-connecting is done which sometimes uses the Institute’s performance guidelines in vain.

Now with that said, I’m not suggesting that the new release of the document will prevent further abstractions of the Institute’s guidelines.  In fact, the bulk of the new release is very much a rearrangement of the Institute’s earlier text.   The rearrangements and new additions however, represent a clarification and stronger position statement of the features required to achieve a given Tier rating, and a stronger articulation of the rationale behind creating the Tiers in the first place.  This is important, as it should help to avoid some of the temptations by the reader to abstract, as mentioned above… including the temptation to define facilities in terms of “Tier 2.5,” “strong Tier 3,” and so on. 

One of my favorite changes in this new release is the move away from specifying power density in terms of watts-per-square foot.   Instead, this release of the Institute’s document takes the approach of kilowatts-per-cabinet.  This is much more relevant to the way we’re planning data center space today, and reflective of the needs of higher density data processing equipment (i.e., blades), and of popular dual/quad processor demands.