This is the first product to intelligently use flash memory, according to IO Turbine. The company says flash memory is more expensive on a per-gigabyte basis but is orders of magnitude lower in cost per input/output operations per second (IOPS). The company says that its solution delivers 10 times or more performance improvement by identifying the highest priority data and offloading IOPS from primary storage to flash, delivering performance directly to designated virtual machines.
Currently available in beta, Accelio provides a significant alternative to buying more drive spindles or solid state drives (SSDs) to address the IOPS bottleneck, says IO Turbine. It says that for every 10,000 IOPS that can be offloaded from storage to flash, users can reduce the number of spindles by 50 to 100 disk drives.
“IO Turbine’s use of its Accelio Software to increase IOPS performance is a clever and useful approach,” says analyst David Hill, Mesabi Group. “The question, however, is for what use cases is their approach compelling, and for which use cases are existing competitive solutions good enough? That remains to be determined.”
SSDs bring something to the party that's not well-aligned to existing system architectures, but Accelio is designed to help work around this problem, says Jim Handy, semiconductor market research, Objective Analysis. “The problem is that computers view storage and memory as two entirely separate entities, and virtual machines want for all storage to be shared. Meanwhile, the fastest SSDs are those that are not designed to be shared, but rather attached, DAS [direct-attached storage]-style, into a server. By helping to hide this incongruity, Accelio should be able to match these SSDs' ultra-high performance with virtualized systems.”
Handy agrees with IO Turbine that, in addition to large enterprises, this software should be attractive to server OEMs. “It also opens up a new market for PCIe SSDs. We already know that the PCIe SSD market is very hot. This heats it up even more.”
Just how much improvement is needed in performance, Hill asks. “If high performance is everything, a DRAM-based approach, such as offered by Kaminario, may be the answer. If simply improving cost/performance in a storage array does the job, then using flash as Tier 0 in a disk array may be enough. Then again, host-based solutions may be a solution or even a hybrid of host and disk array use of flash. In other words, other vendors will have to think carefully before deciding to include an IO Turbine-like approach in their flash mix.”
See more on this topic by subscribing to Network Computing Pro Reports Best Practices: Maximizing I/O (subscription required).