Recapping 2011, Predictions for 2012

Very early in 2011 I had the opportunity to meet with top storage industry veteran who debated with me whether there would ever be a significant solid-state storage appliance category. His view rather was that mechanical storage would remain the focus, but be supplemented with a modest amount of flash cache (think cheaper DRAM).

In stark contrast, Gartner recently published a report on Emerging Technology Analysis: Enterprise Solid-State Appliances (November 2011), which states “The [all] solid-state appliance market generated only $95 million in revenue in 2010, but Gartner believes this market will grow to $3.96 billion in 2015.”

Why the big change? Several factors in our humble opinion:

  1. Mobile devices like smartphones and iPad continued to prove out the reliability of flash to the masses;
  2. With the IPO of Fusion-IO, the industry got broader visibility into a handful of large public cloud flash deployments;
  3. TPC benchmarks based upon ultra-high performance all flash appliances continued to accumulate; and last but not least,
  4. Pure Storage demonstrated that high performance deduplication and compression could provide >5:1 reduction for typical database and virtualization workloads without compromising sub-millisecond latency.

#4 is significant in that we estimate 5:1 data reduction brings the MLC flash used by Pure Storage within our array in cost parity with hard drives in mechanical storage arrays.

So in retrospect 2011 was a great year for flash in the data center, but 2012 is going to be even better. Here are three fearless predictions:

  1. High-performance deduplication and compression will become mandatory for all-flash storage appliances just as they did for disk-based backup solutions. In the early days of Data Domain, the motto was “Tape sucks. Move on.” Inline data reduction allowed Data Domain and their competitors to offer the media customers wanted (disk) at the price point of the one they didn’t (tape). Well, for performance storage, flash is the new disk and data reduction done right allows all-flash solutions to be price competitive with mechanical disk.
  2. The emergence of sub-20nm flash lithographies will make it much harder to build storage systems out of raw flash. SSDs with ASICs and exotic signal processing will be king. Apple likely snapped up Anobit in recognition of this. In 2011, Pure Storage formed strategic partnerships with Samsung and STEC to ensure we would be well positioned for the next generation of flash technology. More on this soon.
  3. Purpose-built flash storage arrays will be proven out to be no-brainers: >10X faster, >10X more power efficient, as well as more reliable and less costly than the mechanical disk alternatives. Disk is the new tape. Consider the perspective of a CPU doing random I/O. Over the past couple of decades, CPUs have become hundreds have times faster, while hard drives have gotten ~1000X denser, but their mechanical performance has been flat. It’s as if we made a bus 1000X longer, but didn’t add any new doors. And thus our ability to get information on to and off of hard drives randomly has been exponentially diminished. (To give credit where credit is due, Jim Gray was—as usual—well ahead of the rest of us. Check out his 2006 PowerPoint talk entitled Tape is dead, Disk is tape, Flash is disk, RAM locality is king).

Although the industry is now increasingly in agreement that flash will redefine the enterprise storage landscape, there is still much that is debated around what exactly is the right architecture for flash in the data center. We think we at Pure Storage have a compelling recipe for addressing the technical and economic hurdles to broad flash adoption, and hope you’ll take a closer look at our FlashArray architecture to learn more. From all of us at Pure Storage, Happy 2012!