Like reliability, energy efficiency has to be designed into the development of these systems. If not, they can consume as much power as the legacy systems they are bought to replace. The end result would be a data center with plenty of physical space but no power to add any more components. This would lead to the strange request for a new data center even though the current one is half empty. To prevent this outcome manufacturers have to add power efficiency to every component possible. 

The logical first step for manufacturers, but unfortunately only a step, is the implementation of MAID (massive array of idle disk) technologies that spin down or power off drives when not in use. The challenge is that most manufacturers offer two operating modes when it comes to MAID, ‘active mode’ or ‘power saving mode’. These modes can range from a spun down but ready state to a totally powered off state with the drives parked, which means a slower response to application requests. The drawback is that manufacturers are making users choose between sacrificing power savings for a better response time or response time for better power savings.

What’s needed is a series of intermediate steps that gradually send the drives to their most power-efficient state as Nexsan does with its AutoMAID systems. This typically starts with the first level being the normal full speed access for when a volume is being written to. Ideally the next step would be to have the disk drive heads parked, reducing power consumption by about 15%-20%, but keeping a sub-second access time to the first I/O. The next step would be, in addition to parking the heads, to slow the platters 30%-50% from full speed. This would bring another 20% or so power savings but only reduce time to first I/O by about 10 to 15 seconds. The final stage would be the classic MAID setting, platters are stopped spinning, heads are parked and the drives are put into a sleep mode. The effect of the final stage could be as much as a 60%-70% reduction in power consumption. 

Storage managers no longer have the time available to them, (if they ever did), to manually make adjustments to their disk array systems. These storage systems should automatically progress through the different levels by implementing intelligence on the array controller that can dynamically monitor activity to the volumes it manages. After a user adjustable period of time the automatic MAID system should begin the progression though these various power saving steps. 

Without this granular MAID capability, most systems will never be allowed to reach the point of that final MAID stage, especially when it applies to their use as primary or secondary storage. With this granular nature there could easily be times even during the business day and certainly during off hours, when some power conservation could take place. Also, the automation and this power/performance granularity should make adoption of MAID less risky. Since there is limited performance impact (less than a second) in the first phases, there’s less risk in enabling MAID. Many storage managers should be surprised at how often their storage enters the stage one condition, each time creating a 15%-20% power savings.

Obviously, MAID is only saving power when the drive is not being accessed. When efficient storage systems are used for either primary or secondary storage, chances are higher that they will be powered on at full speed. While the multi-stage MAID technique described above will help and should be a requirement, manufacturers of efficient systems need to do more. These space efficient systems also need to be power efficient while they are being used.

The first step is to use better, more power-efficient components. This is a challenge for some manufacturers because, while the efficient storage system may consume less physical space due to denser packaging and higher capacity drives, it often uses nearly as much power as does the legacy system it was designed to replace. The reason is that space-efficient arrays often contain a similar number of disk drives, just occupying less space. And since they use the same drives, they can have the same wattage requirement as do the traditional ‘non-space efficient’ storage system.

Few vendors, like Nexsan, will go the extra step of implementing power efficient components because it adds to the upfront cost of the system, even though the IT department may end up spending many more times that cost over time, powering and cooling these less efficient components. When it comes to component selection it’s important to remember that for every watt saved in internal power consumption, almost three external watts are saved because of power delivery inefficiencies and cooling. 

For example, power supplies in a MAID environment often still run at full power, even when disk drives are in a reduced power mode. These generic power supplies simply shunt the excess current to a resister. If they were to select a more sophisticated power supply, one designed for this type of environment, the result would be a net reduction when MAID drives are engaged. Even when the disks are active the power supply should be able to monitor usage and temperature and adjust fan speed as needed. Again, as these adjustments take place, more power efficiency is derived.  

Lastly, by addressing component design, manufacturers can potentially attain some other power benefits. Through the use of ASICs and custom-designed controllers, array efficiencies can be improved, something not possible with generic components. Similar to how a Quad Core CPU processor uses less power than four single core processors, a custom designed controller with less chip count will use less power. An important byproduct of this is also improved performance and reliability since there is less inter-chip communication required.

The final aspect to consider in power efficient design is how the system handles heat dissipation. Disk drives are, of course, the number one component that fails in disk array systems and heat is generally the culprit. This is especially true in space-efficient systems where drives are packed very close together - heat dissipation is a priority. Unfortunately, without proper design, cooling can also consume a lot of power, either internally with large fans and power supplies to drive them or more often externally, with more air conditioning in the data center. In either case, additional power is consumed to keep these systems running at the correct temperature.

Air flow is critical in these systems. Proper air flow can decrease power consumption as well as improve drive reliability. The key to air flow is pressure. Storage system manufacturers need to be able to channel air flow through tight, more directed spaces allowing air to be used efficiently. Many though, just randomly blow air through the unit wasting this valuable asset. By designing systems that can properly move air, with pressure through the unit, the cost to cool the storage system comes down, both internally and externally.

Making efficient storage systems that consume less space but use the same amount of power will lead to data centers that run out of power before they’re full. Applying energy efficiency to these space-efficient systems will allow for the best of both worlds; a densely packed data center that services the needs of what previously took two more data centers. Building energy efficient storage systems that are also space efficient requires moving beyond basic MAID to a more advanced technology that will allow MAID to be used for more than just backup and archive data. In addition, storage system manufacturers need to design in energy efficiency by leveraging specific energy efficient components and creating systems that can properly move air through the chassis.

Nexsan is a client of Storage Switzerland

George Crump, Senior Analyst