Quocirca has been talking with a number of storage vendors over the past few months. All of them are talking about the part flash memory will play in their portfolios, from the use of hybrid flash/magnetic drives, through top-of-rack tiered SSDs to all flash arrays and flash-based PCIe in-server cards.
One vendor of interest has been Violin Memory. According to Gartner, Violin hold the top position in sales of flash-based storage systems with 19% of the market. However, it is still a relatively small market—but it is growing fast. The question is—can Violin keep its position as THE leader, or even as A leader?
In a previous blog piece, Quocirca discussed many of the different approaches as to how flash is currently being used by storage vendors. The main usage is based around the use of SSD—solid state drives, which are essentially a plug-replaceable alternative to spinning magnetic disks. Storage vendors apply some extra 'smarts' around the resulting system to deal with the specific issues around using flash, and the end user gets a storage system that is several orders of magnitude faster than they had before.
This is good—but as flash-based systems become more the norm, then it may not be enough. For those who can remember back far enough, the replacement of tape (a serial storage device) by disk (a less serial device) was viewed with great joy by everyone concerned—at the time, the difference in performance was again several orders of magnitude. Now, squeezing every last ounce of performance from disk is showing that it is reaching the end of its high-power life: flash is the future.
Violin is one of small number of vendors which has decided to take a more radical approach to flash storage. Rather than take an SSD approach, it has built a system from the ground up to make the most of flash: no disks—but it still appears as a standard disk-based system to the outside world. LUNs can still be used, backups and restores can still be applied. However, the systems are engineered to optimise performance from current flash memory technology.
A further blog piece will drill into the differences between different flash memory types, but Violin, along with other flash storage vendors such as Pure Storage, are focusing on the cheaper multi-level cell (MLC) memory type (as opposed to the faster single level cell, (SLC)). Nominally, MLC cells are slower and suffer from considerably less longevity through the erase/write process than SLC. To get around this, vendors have to apply intelligent management to minimise these issues.
Violin does offer SLC-based systems, but its view is that while there is a place for both SLC and MLC, in the majority of cases its approach to optimising the performance and endurance of MLC will meet the majority of needs. The compelling use cases for the extra speed that SLC provides are in verticals such as financial trading and other applications where real-time processing is needed. The view is that the speed improvements that customers will find in moving from spinning disk to any form of flash memory will be massive. Only once the use of flash storage has bedded down will any further improvement in performance from MLC to SLC be noticeable and lead to further mainstream usage of SLC.
At the moment, there are few manufacturers of the base flash memory chips. The main ones are Micron, Samsung and Toshiba—this is an area where Violin says that it is differentiated. Whereas the majority of other flash storage vendors take the SSDs from one of these vendors and build a chassis and 'smarts' around the disks, Violin has a deep technology agreement with Toshiba. This enables Violin to have greater vision in what is happening with Toshiba’s flash subsystems and to create storage arrays that are not built from SSDs, but from its own memory modules, which it calls VIMMs.
Even IBM, with its acquisition of Texas Memory Systems (TMS), depends on Samsung for its flash memory and does not yet have the power to have the deep insight into flash futures that Violin currently has. However, IBM also believes that the future for flash-based storage is an engineered system—and Quocirca expects that the IBM/Samsung relationship will move to be more strategic over time.
Flash-based storage systems provide other benefits beyond raw speed. Better storage performance can improve CPU utilisation. Violin’s figures show that a typical real-world customer can achieve a server consolidation of around 75% in a like-for-like situation, replacing existing spinning disk storage with flash-based arrays. This not only means that server costs are lowered, but also operating system and application licencing, power and cooling and real estate costs are lowered appreciably.
However, moving to flash is unlikely to be a single, fork-lift move for most organisations. How to best tier older spinning disk with flash-based storage, whether SLC should be part of the mix and whether PCIe-based in-server cards should be used will still be a point of contention where professional skills will need to be applied to get the most from such a hybrid environment.
The fact remains, though: if you do not have flash storage on your radar now, then it is likely that your competition has. If the performance of its web site increases by orders of magnitude and if the manner in which it deals with its main business processes is fundamentally transformed, where does that leave you?
Up the Swanee River without a paddle, probably—blowing a Swanee whistle, rather than playing a Violin.