Keeping the Lights On � Literally
The Uptime Institute reports that 1U servers are consuming around $700/year worth of power and this cost is increasing at a time when budgets are either static or being cut. If left unchecked, power consumption will be funded at the expense of more important initiatives.
In addition to cost concerns, some data centers managers are finding they cant even get additional power from their utility and are essentially constrained to operate within the limits of the power feeds they currently have. Combine the two and you can readily see that unless power demands are managed, IT will be limited in terms of supporting the business.
To begin curtailing power demand, the following are recommended areas that groups review:
Utilization and Efficiency - In general, as the utilization of a server, computer room air conditioner (CRAC) or UPS increases, then the efficiency of the units operation increases. This basic concept can then be applied in a number of ways including the need to better plan capacity, as well as the next two points we will touch on.
Find and Remove Ghosts - It is not uncommon to find abandoned servers that are no longer needed but are still working. In data centers the Uptime Institute looked at up to 30% of the servers were ghosts. If you factor in the power and other support costs of these servers, sitting idle means the organization may be spending a fair amount of money with little to no benefit.
Consolidate and/or Virtualize - If servers are only being utilized 6% to 15% of capacity then they are operating inefficiently in terms of power consumption. First, unnecessary IT services or applications that are no longer needed should be decommissioned. Next, by either consolidating applications onto existing servers and storage systems or by using virtualization technology, data centers can reduce the number of discrete servers and thus power demands.
(Note: virtualization and the journey to high-density computing is one of the most important efforts medium to large IT organizations can embark on. But, planning must be done to ensure that the data center can adequately support the power and cooling demands of these systems.)
Address Air Flow - IT equipment needs to be cooled in order to reduce the chances of component failure and premature aging caused by temperatures outside of design limits. As a result, data centers use a variety of means to cool IT equipment. In many data centers the air flow hasnt been properly maintained to optimize cooling.
The use of hot aisle/cold aisle designs, leveraging economizers, partitions to limit commingling of cool intake and hot exhaust air, proper placement of ducts and cooling tiles, blocking uncontrolled airflow by using panel blanks and seals in racks will all help improve the efficiency of cooling and thus the power demanded by those systems.
Power Management - Organizations need to carefully study their resource demands and then configure device power management options accordingly. Despite most organizations seeing changes in demand levels dropping during off-peak hours, the power demands are staying static and this indicates that power management capabilities arent being effectively used.
Recognize CapEx/OpEx Trade Offs - Given that many organizations have either frozen capital investments or are at least carefully scrutinizing capital investments, management must carefully understand the tradeoffs between capital investments and changes in operating expenses. All things being equal, newer IT equipment makes more efficient use of power.
While capital investments are typically scrutinized, operating expenses around power often are not. Management needs to understand that to freeze investments in newer more efficient systems will result in higher operating costs. Already we are seeing IT assets that incur four to five times their purchase costs in the form of operating costs over their life. Sooner or later, the organization will pay in one form or another.
Monitor Power Usage - IT needs to implement the necessary monitoring systems to understand how power is being consumed in the data center. The ideal situation is one where IT can tie power consumption to business services, understanding cooling per service, where power is lost, and so forth. The intent is to collect meaningful data that can be used in management decisions and process improvement.
In closing, there is a tremendous amount of activity in the trade press around data center power, linking it to green IT and environmental impacts. There is nothing wrong with this connection as it is very real: data centers really are consuming massive amounts of power and IT needs to take the necessary steps to gain control.
George Spafford is a principal consultant with Pepperweed Consulting and a long-time IT professional. George's professional focus is on compliance, security, management and overall process improvement.