Pure Storage came out of stealth with its Pure Storage FlashArray FA-300, which the startup’s executives hope will bring “flash to the masses” with a price point of $5 per gigabyte of usable storage.
Pure Storage’s FA-310 and FA-320 contain 2.5 inch SATA-based MLC drives. The FA-310 is a single controller system that supports 5.5 TB or 11 TB of raw capacity, while the FA-320 has two controllers and from 11 TB to 22 TB of raw capacity. Assuming a 10-to-1 deduplication ratio brings the maximum raw capacities to 110 TB for the FA-310 and 220 TB for the FA-320.
Texas Memory Systems added an enterprise MLC (eMLC) system to its lineup of all-flash storage devices. The TMS RamSan-810 comes in at approximately $13 per GB, roughly half the street price of its single-level cell (SLC) RamSan-710. The 10 TB RamSan-810 can scale to 400 TB with a performance of 160 GBps and 12.8 million IOPS in a 40U rack, according to TMS.
A new face in flash storage
Newcomer Pure Storage joins a small group of vendors selling all-SSD arrays, including TMS, Nimbus Data Systems Inc. and SolidFire. Pure Storage uses SSD drives from Samsung, a strategic investor that has helped the startup amass $55 million in funding.
The FA-300 devices have active/active clustered controllers for high availability, with two six-core Intel Westmere CPUs (12 cores) per controller or four CPUs (24 cores) per array. Each array has 48 GB of DRAM per controller, plus additional non-volatile RAM (NVRAM) to persist data in the event of a power loss.
“The joke around here is that CPUs are cheaper than flash,” said Matt Kixmoeller, Pure Storage’s vice president of product marketing/management, referring to how the FlashArray uses CPU-heavy deduplication software to reduce the cost of flash. “Our focus is all-flash and we believe we can deliver flash at under the price of a hybrid [flash and hard drive] solution.”
The FA-300 series is in its fourth beta testing release, and will be generally available by the end of the year. Final pricing will be set closer to release, Kixmoeller said, although Pure Storage has tipped its hand with its $5 per gigabyte claim.
The $5 per gigabyte cost (which includes high availability) takes into account hardware, software and assumes a 5-to-1 data reduction, said Michael Cornwell, Pure Storage’s director of technology and strategy. The company plans to come to market with a much cheaper price point than other all-flash array companies. Nimbus Data has a cost per gigabyte at just over $12.
Unlike Nimbus Data, which keeps costs down partially by manufacturing its own SSDs, Pure Storage gets SSDs from Samsung for its arrays. However, Cornwell said Pure Storage will keep its price low because of how it manages data on the flash drives with its Purity operating system. The Purity software handles inline data reduction through global deduplication, compression and thin provisioning. Also, the software sends I/Os in parallel to the SSDs to improve performance and refreshes the data on the cells to keep the electrons stable so that data doesn't fade.
“We're able to use MLC in an enterprise environment because of how we manage the flash,” Cornwell said. “Our software is always refreshing the data so you never experience that event. We're in the background moving data to different cells to ensure the data doesn't fade away.”
Pure Storage also offers data protection through its proprietary RAID-3D. “We optimize RAID for flash,” Cornwell said. “Flash doesn't have the issue of locality of data that disk has. We're able to lay out data in a hierarchical manner, which enables better performance of the data store. The way we lay out the data allows for fast rebuild times.
Nimbus Data and TMS call their flash eMLC, while Pure Storage and Samsung label theirs MLC. Until recently, MLC was considered good only for consumer devices while SLC was the choice for the enterprise. The high price of SLC has limited its adoption, though, prompting flash vendors to tweak MLC to improve its reliability and performance.
Mark Peters, a senior analyst at Enterprise Strategy Group, said there's not necessarily a difference between eMLC and MLC. “The ‘e’ of eMLC is to convey that you have the software in the system to overcome some of the inherent challenges with MLC,” he said. “Pure Storage has management software on top of the MLC. They're just not calling it eMLC.”