There’s no question that RAID is ubiquitous today; the backbone storage technology in modern day data centres. It’s likely to remain so for the foreseeable future with so much cloud computing being delivered by data centres using RAID technology and by the massive amounts of data being generated by phenomena such as social platforms, smart clients and mobile Internet devices, big data, and the proliferation of video across consumer and enterprise platforms.
The adoption of enterprise flash-based solutions is the next logical evolutionary step in storage technology as the gap between data growth and IT infrastructure investment has widened over the years, creating (performance or I/O) bottleneck problems for mission-critical applications.
So how can organisations bridge these gaps? For many, solid state drives are well-placed to close the gap between compute and storage performance because they have a much faster random access time and data transfer rates, lower latency and consistent read performance compared to traditional hard disk drives. However, the cost of replacing the entire storage infrastructure with SSDs is impractical and cost-prohibitive, so a more economic option is to integrate flash technology into enterprise system architectures to deliver significant improvements.
The evolution of RAID
Over the years, a number of standard RAID schemes have evolved, known as levels. RAID 0 improved performance and added storage but had no fault tolerance while RAID 1 enabled mirrored data to be written identically to two drives. RAID 2 and RAID 3 synchronised disk spindle rotation and stored sequential bits and bytes on a parity drive.
With RAID 4, files were distributed between multiple drives that operated independently, allowing I/O requests to be performed in parallel. But with all parity data stored on a single drive, it could suffer from a performance bottleneck. RAID 5 distributed parity along with the data. If a failure occurs, subsequent reads can be calculated from the distributed parity. RAID 6 provides fault tolerance of two drive failures, making larger RAID groups more practical for high availability systems.
Distributing data across multiple drives in a RAID system can be implemented at a software level or a hardware level. At the hardware level, RAID controllers can support many operating systems because they present the RAID as another logical drive.
They include a read/write cache, so they can improve performance and as the read/write is non-volatile, pending writes are not lost in the event of power failure, as long as the cache is protected by a backup mechanism. Hardware RAID provides guaranteed performance and does not add computational overhead to the host computer, but because controllers use proprietary data layouts, it may not be possible to work with controllers from multiple suppliers.
Bridging the gap
So far, the evolution of RAID has generally managed to keep pace with the demands placed upon it, but for how much longer? According to research from Cisco, network traffic is likely to experience a compound annual growth rate of 32% between 2011 and 2015. IDC predicts storage capacity will need a compound annual growth rate of 50% in the same period. These trends are emerging at the same time as Gartner has estimated the compound annual growth rate for IT spending is 5% and the figure for telecom equipment spending is 7%.
With the amount of data being stored and accessed worldwide growing at a massive rate, a performance gap is emerging as RAID struggles to keep up with performance acceleration on the processing side. It is common for CPUs to achieve write speeds of 1 nanosecond to L1 cache memory, 10 nanoseconds to L2 cache and 100 nanoseconds to main memory. This is significantly faster than the 10 millisecond write speed to tier 1 storage, or the 20 milliseconds to tier 2 storage and to near line storage. This represents a latency penalty of 100,000x for leaving the memory hierarchy.
The combination of this performance gap and the explosion in data growth and network traffic is likely to strain RAID storage infrastructure, creating bottlenecks, throttling application performance and making it harder for companies to extract the full value from their data. This is of even greater concern in a world where speed of access to data is of the essence and everybody wants access to everything immediately.
Redefining data centre performance
Using flash-based storage with existing storage can save significant amounts of money because tiered storage arrays with flash on the RAID controller (combined with intelligent software) can replace a huge amount of disk drives that are otherwise necessary to maintain I/O rates in traditional storage arrays. The combined approach enables businesses to intelligently use flash storage and their existing hard drives together in a way that can give them optimum price/performance in a tiered storage environment.
Combining PCIe flash technology with intelligent caching and management software can deliver an impressive performance acceleration that depends on configurations and application but is easily a large multiple of the HDD performance. It is not unusual to experience application performance acceleration of 5-6x but some cases have been reported up to 30x.
Solutions can be delivered with different capacities to suit different requirements. High capacity PCIe flash solutions can be used for primary storage to deliver high value non-transparent storage, but have a cost premium attached. Medium capacity flash solutions that combine PCIe flash technology with intelligent caching software can accelerate high value SAN and complex DAS connected storage and provide a balance between cost and value. Low capacity flash solutions combine a RAID controller card with on-board flash and intelligent caching software to accelerate DAS connected storage and bring the value of flash to the masses.
The intelligent deployment of flash technology can help RAID evolve to meet the challenges presented by the ‘data deluge gap’. And perhaps it’s fitting, given the preoccupation with performance that informed the publication of ‘The Case for Redundant Arrays of Inexpensive Disks’ in 1988, that flash technology is also being deployed to help RAID address I/O performance issues.