Airflow Management Considerations for the Colocation Tenant12 min read
Last month we talked about airflow management considerations for the colocation data center landlord. Key issues revolved around profitability and competitiveness, with important considerations in service-level agreements and billing models and additional growth, density and education considerations. Billing models and service level agreements will also be key determiners of airflow management tactics and investment strategies for colocation tenants, but with an occasional different twist on perspective.
Obviously, if the billing model is tied directly to kW hours then it behooves the tenant to optimize data center airflow management on all levels, conveniently defined by Upsite’s 4Rs (raised floor, rack, row and room) with all the elements, structures and accessories to maximize separation between supply air and return air and thereby minimize bypass airflow and recirculation airflow. After all, effectively implementing all four R’s of airflow management versus not addressing it all can easily make a $700,000 annual difference in billing for a 1 MW IT load or a $3500 difference for a 5 kW rack, putting this airflow consideration in that “no-brainer” realm. Not so fast, my friend. Did I say obviously behooves? What if you are in a multi-tenant building and all the other tenants seem to own stock in the utility company and have all their tower servers sitting on shelves on open racks? Or the tenant next to you has a rack full of side breathing switches blasting hot air into your “cold aisle”? Or the landlord is using a thermostat strategy for temperature control and has established a 70˚F (or 69˚ or cooler) set-point? Or all the landlord’s air movement machinery has single speed fans? And the exception for a side of beef hanging in the corner of the data center should go without saying – but I said it anyway. The point here is that while maximizing the effectiveness of all elements of airflow management can pay big dividends when a colocation tenant is being billed for straight kW hours used, we cannot lose sight of the fact that airflow management does not automatically produce energy savings; on the contrary, airflow management is an enabler of all those temperature control and fan speed management strategies that directly produce energy savings.
Therefore, where colocation billing is tied directly to kW hour use, there are actually several airflow management considerations for the tenant. First, if it is a single tenant facility and all pumps and fans are on some kind of variable speed control and the SLAs and hardware accommodate temperature control by ICT equipment inlet temperature, then it makes good economic sense to invest in the most effective airflow management hardware and techniques appropriate to the installation. (See recent blog pieces on ideal conditions for hot air containment or cold air containment for further guidance.) If, on the other hand, the kW hour billing model is in a multi-tenant colocation facility where the non-UPS load is apportioned among the tenants according to share of IT load, AND the other tenants appear to be exhibiting relatively mature airflow management discipline, then it would make sense to at least invest in airflow management commensurate with one’s neighbors. If a particular tenant’s share of the total load was half or more of the total multi-tenant load, then it might even make sense to exceed the quality of hot and cold separation of the installed community; payback might be a little longer but the overall ROI should still be acceptable. On the other hand, if it is a multi-tenant facility and the prospective neighbors are essentially lacking any apparent sense of airflow management discipline, then the most appropriate consideration might be to keep shopping for space.
Other billing models will suggest other airflow management considerations. For example, if billing is by the rack, the prospective tenant should look for SLAs that define good airflow management practices and then check the space to ascertain how well those rules are enforced. Best practice-based SLAs and subsequent tight enforcement should mean the space could support higher rack densities and thereby be a more-bang-for-the-buck winner. Without these elements, the tenant is shopping for best price, and I don’t mean just an incrementally best price. For example, $700 per rack per month might seem like a much better deal than $1500 per rack, but if you can get 20 kW in one $1500 rack but it takes four $700 racks to support the same 20 kW IT load, then we’re in the middle of the old “When’s a deal not a deal” deals. Conversely, PUE billing models will have similar considerations to kW per hour billing models. A single tenant facility with a mechanical plant capable of exploiting good airflow management practices will justify investments in the maximum physical separation of hot and cold air masses. A multi-tenant facility with a responsive mechanical plant, good SLA direction and enforcement and a reasonable methodology for apportioning PUE will likewise typically justify investing in best practices airflow management; whereas the absence of all those elements might be a good reason to look elsewhere.
“Airflow management is an enabler of all those temperature control and fan speed management strategies that directly produce energy savings.”
The previous two paragraphs, to even the most casual reader, should suggest there might be some kind of relationship between colocation billing models and SLAs as far as airflow management is concerned. As a matter of fact, it is difficult to talk about one without the other. The most straightforward relationship will naturally be in a single tenant facility wherein the SLA is the result of a one-on-one negotiation. The tenant should look for variable billing by kW or PUE and a mechanical plant that is capable of responding to conditions created by good airflow management, such as supply temperature set points and pressure-controlled airflow. The associated SLA for a mechanical plant without economization will look for supporting a maximum allowable server inlet temperature while allowing for a minimal variation in the minimum server inlet temperature. A 5˚F difference between the minimum and maximum equipment inlet is a conservatively safe number, and that difference can be reduced based on the tenant’s experience with the deployed airflow management solutions. This SLA will drive the facility to reduce airflow and increase supply temperatures to optimize the billing model for the tenant. On the other hand, if economization is part of the mechanical plant, the SLA should allow for the full temperature range for equipment inlets that would be acceptable to the client; for example, they might pick 64.4˚F to 80.6˚F if they wanted to live within the ASHRAE TC9.9 recommended environmental envelope or 50˚F to 95˚F if the wanted to live within ASHRAE’s allowable range for class A2 servers. However, they define that environmental envelope, the SLA should include a provision to run on economizer for all hours capable of delivering a supply temperature to support a maximum equipment inlet temperature within 1-2 degrees of the SLA maximum. These parameters will offer the tenant the most attractive usage billing conditions. Any of the above scenarios will obviously be optimized by the maximum separation between cold and warm air masses.
Conversely, in a multi-tenant facility with an SLA defining the data center temperature within some narrow range such as 70-75˚F or defining a maximum temperature at the cooling unit return air intakes as something in the mid 70’s or (gasp!) lower, then any investment in airflow management would not likely have any reasonable return on investment and might only be reasonably considered as a hot-spot mitigation strategy if, fact, hot spots became an issue. Another consideration for multi-tenant facilities would be whether or not there is an SLA requirement that all ICT equipment breathe front-to-rear, or that non-compliant breathers are housed in support equipment that correctly re-routes airflow. In a multi-tenant facility with SLAs requiring a narrow data center operating temperature and defining a low maximum return air temperature and not requiring front-to-rear breathing or adjusted breathing equipment, it is best to steer clear of kW or PUE billing and go with rack, cage, or footprint billing instead.
There are a plethora of issues involved in selecting an appropriate colocation provider and all I’ve discussed here are some cost issues associated with billing models and SLAs and how they bear on airflow management considerations. For this very narrow criteria, it really is not a very complicated picture. For usage billing landlords, we should see a mechanical plant and SLAs that will allow effective airflow management to deliver the maximum value to the tenant. Otherwise, we either want a different billing model, different SLAs or a different landlord.
Data Center Consultant
Let’s keep in touch!
Airflow Management Awareness Month
Free Informative webinars every Tuesday in June.