Archive as a strategy should drive down the cost of storage and it should be a top priority in 2009, as IT budgets get tight. Along with driving down the cost to store data, the ability to drive down the power and cooling costs of that storage should be a critical consideration in any archive strategy.

The importance of this combination has not been lost on traditional storage manufacturers. Almost every storage manufacturer has announced a technology based on power efficient drive technology. With this announcement comes the claim that they are in the MAID (Massive Array of Idle Disks) space, joining the market of pioneer Copan Systems. To develop a system that is power efficient requires more than just the ability to put power efficient drives into the storage shelf.

What are green or power efficient drives? They are drives from the hard drive manufacturers that essentially have the ability to accept commands to spin down or even power off. The goal of systems that will use these drives is to reduce power consumption of the storage system. The strategy is to move old data to these drive shelves and then to hope for infrequent access so the drives may spin down and deliver that power efficiency. Essentially a long-term disk based archive.

The challenge facing manufacturers new to the power efficient storage market is to obtain maximum scalability and density with minimum power efficiency requires more than just inserting power efficient drives into the existing storage system. To achieve this goal requires an entire eco-system of drives, system hardware and software intelligence.

Optimizing Storage Hardware for MAID

With the use of power efficient drives, changes can and should be made to the hardware to take maximum advantage of those drives. In most systems that allow for power efficient drives the shelves are “fully” powered. This means they can not be powered down, or require less power. They must be ready for all the drives to power on or spin up simultaneously. There is a lack of intelligence in data placement so that accessing one file from the disk archive can result in the entire shelf of drives to become active.

Not only does this require a full dose of power but also it needs the same cooling capabilities as a traditional system. Drives cannot be densely packed into the shelf for fear that they would all power on. Which would require a larger power supply or the intelligence to only power on a percentage of the drives. Today these capabilities are lacking in traditional systems.

With a properly designed MAID system data can be placed on the drive that only a finite number of the drives in the shelf actually power on at any given time. This allows for the two big advantages. First, more drives can be packaged into the storage shelf and second less power has to be sent to the shelf on a continual basis. The result is less power utilization on a per shelf basis and fewer overall shelves. Each shelf that can be eliminated because of dense packaging is a significant power saver.

Dense packaging of drives also requires additional thought when engineering the system itself. Even in drive shelves that pack the standard amount of drives, vibration is a problem that causes many drives to report failures. In a tightly packed system it could get worse. Properly designed MAIDs can eliminate this problem all together by having intelligence in the storage software to make sure that two drives side-by side seldom power on at the same time and by counter mounting the drives so they are not rotating right next to each other. Taking advantage of the technology of power efficient drives and proper engineering of the drive shelf makes dense packaging of drives a reality.

Dense packing of drives and vibration also highlights the key differentiator in Power Off vs. Spin Down drives. Drives that are powered off require no power, send out no vibration and as a result can be very densely packed. Drives that are called spin down typically do so by slowing the RPMs of the drive so they require less power. The problem is that they are still vibrating and as a result, dense packing of those drives are more problematic. In a MAID system if you can not densely pack drives then you need more space per drive per shelf, you will need more shelves and as a result your power, cooling and space utilization are not as optimized as they should be. While the wattage difference between spin down and powered of drives is not significant per drive even a 1-watt difference across 1,000 drives quickly becomes significant. Add to this the equipment of additional shelves and power for those shelves and a spin down drive quickly becomes inefficient.

Optimizing the Software for Green Hardware

Intelligent use of the hardware is just half the issue. The software that manages the MAID is also critical. A few of the key points have been touched on. It is the software that lays the data out in such a fashion as to make sure it is unlikely for an entire shelf to be powered on simultaneously for example. Even if multiple drives in a shelf are needed, that intelligence can safeguard against attempts to power too many drives up at the same time, preventing a power supply overload or the need for oversized power supplies that consume too much energy. This can be implemented by only spinning drives up one drive at a time to ensure system safety and power efficiency.

Another component in data layout is how data is written to the system. Can it be laid out sequentially as opposed to randomly? Random writes cause data to be cast everywhere and may cause more drives to be powered on in the event of a read of older data. Also drive count in array groups needs to be small, ideally 3+1 (3 data drives, 1 parity). Using green drives in standard drive shelves leads to traditional larger drive count array groups because of the complexity of managing so many array groups. A properly designed MAID system should be able to manage the multiple groups, presenting a simpler appearance to the administrator.

Data placement becomes a critical capability of the MAID system. It requires an understanding of data access patterns. The system should try to keep active data in an always-on region and then move older data to less frequently powered regions. Whether using spin down drives or powered off drives, there is a power cost when having to bring them to a ready state, sometimes double the active state. Without proper management of how data is laid out on the system, power savings will be quickly lost by the constant powering on and off of drives.

The final and possibly most important role of the intelligence on a MAID system is assuring data integrity. There is a misconception that powered down drives have a lower meantime before failure (MTBF). Studies have shown this not to be the case, in fact with proper drive verification the MTBF of powered down drives actually improves. A key component is the ability to on a monthly basis test the drives for data integrity. If a failure is found then the valid data blocks can be re-distributed throughout the system before data loss occurs. Once again lack of vibration plays a key role in maintaining drive health. Reduction in vibration causes a dramatic reduction in drive failures.

A disk based archive is supposed to reduce the storage line item of the IT budget by reducing the cost of investing in tier one storage. The potential to be able to do so while dramatically reducing power consumption and cooling costs provides a trifecta in budget savings that can allow progress on another project that might have otherwise been delayed.