Fusion io Optimizes Server Flash from Application-Level Viewpoint

Fusion io, which sells flash memory cards for servers, has unveiled its latest product series that doubles data storage capacity of NAND flash and features a slew of enhancements designed to accelerate application performance beyond the basic fast performance of flash as a storage medium.

A new generation of NAND flash technology comes out about every 18 months, doubling capacity of the medium. Fusion io’s new Atomic series represents the most current leap. Gary Orenstein, the company’s senior vice president of marketing, said the new platform brought top mainstream NAND flash device capacity from 3.2 terabytes to 6.4 terabytes.

Capacity cost does not paint full picture

As the cost of flash inches down, it has become an increasingly popular alternative to spinning-disk hard drives in data centers, especially for storage used by applications that require fast delivery of data. Instead of focusing on cost per unit of storage capacity, however, Fusion io promotes other valuable aspects of flash.

“We try not to measure it by capacity points,” Orenstein said. “Flash isn’t so much about storing data as it is about delivering data.” Looking at cost of storage capacity alone means to ignore the main value of flash, which is high performance.

Many vendors today, Fusion io included, work to increase performance of flash beyond the sheer performance characteristics of the medium itself. They do it through technological innovation in management of the storage systems.

One example of such a feature introduced in the Atomic series, for example, is flash-aware compression. While common in traditional storage systems, data compression is not usually employed when flash is involved, because it reduces performance and defeats the purpose of using flash.

Flash-aware MySQL compression

One of the applications Fusion io has optimized Atomic for is the open source database management system MySQL. When data is erased from NAND flash, it is erased from blocks of cells simultaneously instead of erasing from individual cells. To create those blocks, data needs to be moved around. MySQL also moves data around for its own compression purposes.

These processes taking place simultaneously on the same storage area cause the degradation in performance, Orenstein explained. Flash-aware compression exposes the work happening at the NAND device level to MySQL operating at the application level and facilitates coordination between the two. With such intelligence sharing, the impact of compression on performance becomes insignificant, he said.

Tight SQL Server 2014 integration

Another example is work Fusion io has done to optimize Atomic for Microsoft’s latest SQL Server 2014, launched in April. The application has a feature called buffer pool expansion, which expands the amount of memory available to the database to process things in real time. Atomic integrates with this feature, enabling it to expand the buffer pool to flash memory.

Such application-level integration demonstrates Fusion io’s efforts to look beyond working just with the hardware vendors that sell its products as part of their offerings. The company works closely with software vendors, such as Microsoft, VMware, Oracle and SAP, to improve performance of flash when used with the common enterprise applications.