Recently I met with the folks at Atrato, a nascent storage innovator based in Colorado’s front-range city of Westminster. Formally known as Sherwood Information Partners, Atrato came out of stealth mode earlier this year and has since developed significant traction in the VOD space with an innovative architecture that is characterized by high access density, high bandwidth, and high storage density.
The Atrato Velocity 1000 consists of a 2U controller and a 3U storage enclosure called a SAID (Self-maintaining Array of Identical Disks). Each SAID uses 160 2.5” drives (320GB & 500GB) exclusively. The innovative and patented drive packaging delivers exceptional cooling characteristics, and the smaller FF drives are surprisingly power efficient. Data protection is assured via RAID 10 or 50 with user selectable single, double or triple mirroring.
Some high spots of the Atrato technology are:
Failure in Place: The architectural concept of failure in place is fundamental to the Velocity 1000 product. This feature simply means that when drives fail they are not physically replaced. Hence the notion of tightly coupling the available user capacity with enclosure life cycle. The enclosure life cycle is the projected time over which the “as sold” raw capacity will be available to an application. The total manufactured capacity has to include an allowance for anticipated sparing requirements over the guaranteed capacity life of the enclosure. Obviously some secret sauce surrounds the Atrato predictive usage algorithms. A similar concept is implemented by Xiotech with its ISE technology.
This approach eliminates the hassle of managing “hot pluggable” drive FRU’s. Failed drives are isolated but remain in place. Recoverable drives are reintroduced into the storage pool as are partially failed drives, reclaiming the remaining usable capacity. No external maintenance action is required; maintenance is automatic.
The Velocity1000 carries a three-year zero disk hardware maintenance. Note, controllers, fans and power supplies are still hot plug replaceable.
Literally a “cool” design, with the elimination of hot swappable drives enabling innovative design concepts in packaging density and airflow (cooling). Those interested in detail should check out the Atrato patents.
FDIR: Atrato is particularly proud of its FDIR design (fault detection, isolation and recovery). The terminology is adopted from the aerospace industry to describe automation for highly reliable and available systems. It continuously monitors component and system health and is coupled with self-diagnostics and self-healing. This anticipatory, preventative, and correcting methodology helps avoid failure induced service interruptions and performance degradation by detecting problems early and initiating transparent recovery processes. The secret sauce would appear to be in the extensive decision logic based monitoring, stress testing, and failure analysis.
One element that held my attention is what they call abstracted spare consumption monitoring - a neat feature which monitors the remaining usable capacity and predicts future capacity availability (i.e. remaining enclosure life) based on historical user consumption rates and drive failure data. This is a useful tool for storage administrators as they plan future storage requirements. The information is available remotely via CLI or web browser.
Performance: Access density is a measure of the number of concurrent requests a particular configuration can handle. Basically it is a function of the actuator-to-capacity ratio and is measured by dividing IOPS by capacity (IOPS/Capacity). By exploiting 2.5” drives, Atrato can pack 1280 spindles, hence a very high actuator density, into a full 42U rack. When optimized for IOPS and bandwidth, Atrato claims an impressive random small block (no cache) IOPS number of more than 100,000, plus a bandwidth of 12GB/s.
That should be fast enough for most high velocity data!
Factoid: Five 2.5” drives have the same volume as one 3.5” drive, hence more actuators per unit volume.
Capacity: How about 640TB (raw) in a standard rack? This is achievable when the solution is configured to optimize capacity, eight groupings of one 2U controller and one 3U, SAID enclosures, fitting into a standard 42U rack. Each SAID has 160 drives packaged in a very innovative configuration delivering an outstanding storage density of over 99TB/sq ft. A quick competitive comparison highlights the Atrato advantage.
- NetApp FAS - 27.5 TB/sq ft;
- Isilon IQ 1200 - 31.9TB/sq ft;
- EMC Infiniflex 12000 (Hulk) - 50.21TB/sq ft;
- HP ExDS 9100 - 67 TB/sq ft;
- Data Direct Networks - 80 TB/sq ft;
Factoid: A 2.5” drive has almost twice the data density of a 3.5” drive.
Energy Efficiency: Conventional wisdom has declared that the path to energy efficiency is through larger capacity drives, but Atrato has a somewhat contrarian view, focused on more but smaller form-factor drives. Its proof of the effectiveness of this approach is a claim, based on a component level (drive) analysis, of 50% to 80% greater power efficiency versus comparable solutions using 3.5” drives. The interesting question is how much of this drive level advantage translates to end-user benefit at the solution level.
Factoid: A 2.5” drive consumes about 1/3 of the power of a 3.5” drive, suggesting that the 2.5” drive delivers higher bandwidth and IOPS per Watt.
Zero Maintenance: I mentioned this earlier, but it is worth underlining. While I have not completed the necessary homework, I think it is safe to say that zero disk maintenance costs for a three-year period should have an impact on the total cost of ownership, and when combined with space and energy savings this solution should deliver an attractive cost/performance ratio.
While the storage density of the Atrato solution is impressive, especially when compared to traditional transactional architectures, it is the uncompromised IOPS and bandwidth performance that separates this solution from more conventional approaches. Currently Atrato is establishing itself in the VOD and nPVR markets, but it should not be long before it finds its sweet spot crossing industry boundaries to solve a much broader set of application problems where very high velocity data access, data movement, and dense data storage are the requirements.
Action Item: Many applications continue to be IOPS bound. Existing storage solutions are unable to deliver the necessary very high-end performance, hence the interest in SSD. However, new disk architectures that are tuned to deliver very high performance, IOPS and bandwidth should be considered before an SSD decision is made. With a claim of 100,000+ IOPS and a bandwidth of 12GB/s, Atrato is a vendor with such an architecture and should be one of the solutions considered.
Footnotes:
