2013 is the year that flash became “standard equipment” for mainstream storage arrays. As the year comes to a close, it’s hard to envision a new enterprise-class array being deployed without flash in some shape or form.
Here’s my flash year-in-review, plus five top tips for IT success in 2014.
Today, flash-based solid state disks (SSDs) are commonly tiered with traditional hard disk drives (HDDs) in hybrid arrays or deployed in the form of 100% all-SSD arrays (aka, all-flash arrays, or AFAs).
Less well known, but equally important, was the widespread adoption of flash as an adjunct to storage controller DRAM. PCI cards filled with flash modules are now able to expand the effective memory cache capacity of storage arrays and further increase performance, providing what is arguably the fastest of all the flash acceleration options. While the same approach can also be applied for server-side flash, that option was not fully embraced by enterprise IT in 2013, perhaps because many infrastructure teams are still unwilling to exchange a few extra IOPS for the disruption involved when installing new cards and software drivers into business-critical servers.
Significantly, there was a relative absence of stories in the trade press indicating problems with flash reliability. This was welcome news given the initial well-publicized concerns regarding flash cell wear-out. It seems that storage vendors have addressed these concerns with wear-leveling software and bad-block management (and that those techniques have worked as advertised).
Prices drop, capacities climb
Throughout the year, flash vendors transitioned their silicon factories from manufacturing 32nm-class flash dies to the smaller and denser 19nm-class dies. This move, along with broader adoption of multi-level cells (MLC) has enabled the current crop of enterprise-class 800GB SSDs, up from the 200GB versions offered just one year ago. Although at least one vendor has been touting their 10nm die, there remains some question as to just how much more capacity can be squeezed out of flash memory in the future.
The economies of scale in flash manufacturing has led to a simultaneous decline in flash pricing as capacities increased. While flash is still far from cheap, enterprise vendors have been able to find creative methods to offset this cost, including the use of lower-cost, consumer-grade flash and experimentation with forward error correction (FEC) to eliminate RAID capacity penalties.
During 2013, no new memory technologies threatened the pervasive growth of flash in enterprise storage systems. However, vendors continued to pursue and promote alternative memory technologies such as memristors, phase-change memory, carbon nanotubes, and spin-torque transfer. While interesting, such efforts have failed to deliver products in significant volume—at least, so far.
Although technologies like phase-change memory and memristors appear ready to go into production, a variety of factors continue to hold up investment. These include high fabrication costs and uncertainty regarding when flash will ultimately hit the price/performance wall and thus need a replacement.
Competing technologies aside, the flash market itself went through its share of pack-shuffling and jockeying for position throughout the year. Of the dozen or so enterprise flash startups that had received significant funding, two were acquired, two announced IPO plans, and two exited the flash business. In addition, the five dominant enterprise storage vendors all revealed product roadmaps with a heavy emphasis on all-flash products.
In recent years, few enterprise storage technologies have seen as rapid adoption as flash did in 2013. Rest assured that the fun will continue in 2014.
What do these changes mean for IT professionals? Here are five suggestions for the coming year:
1. If you are not currently using flash, how secure is your job? Look for ways to implement flash to improve application performance.
2. Think strategically about how to use flash in your storage infrastructure, with an eye towards future requirements.
3. When considering new technologies such as forward error correction—aka information dispersal, or erasure encoding—make sure you fully understand the pros and cons.
4. Tread cautiously when using consumer-grade flash in an enterprise application. Although catastrophic flash failures are rare, the durability of low-grade flash in high stress environments is largely unproven.
5. There are many ways to implement flash, make sure you understand them all. For more, check out my previous post: Speedy storage: Pros and cons of SSDs and flash.
Next up: Big Data in 2013.