The West Coast Offense Arrives in Storage - Executive Viewpoint 2013 Prediction: Convergent.io
Looking at storage innovation these days reminds me of football in the eighties. For years, the offensive game plan was “run to set up the pass”. Offenses were conservative: you ran on first down, ran on second down and if it made sense, you ran again. (Yes, you Don Coryell fans, a few teams went vertical, but that sort of bravado was considered suicidal.)
For a long time, the enterprise IT world was conservative like this. Storage admins knew that they weren’t going to get fired for buying EMC or IBM, or Netapp. CIOs would cringe at the price points and admins would cringe when performance tapered off as more capacity was added to these big box arrays. All the cringing amounted to nothing, it was just part of the enterprise storage game.
And then virtualization came along and exacerbated the suffering. Consolidation ruined the nice tidy application workloads that those arrays were used to serving: by now we’ve all heard of the “IO blender”.
Like a good coach, you could see the problem, but what could you do? -- Storage was the bottleneck. You could buy an expensive array with more disks and bigger RAID sets. You could add caching. You could even buy a pricey all-flash array, but all of these options are like adding a high priced free agent. They are risky and they eat up the salary cap. All of us have a spending cap.
Then came Bill Walsh. With a fast, dynamic and athletic offensive line, an accurate and extremely cerebral quarterback and wide outs who were not afraid of venturing in the middle for short passes, he “passed to set up the run”, and football has never been the same. Bill changed the approach to the game. He used resources differently. He took calculated risks: making effective moves that were surprising only in that they broke from the conventions of how people thought the game was played. He used these three weapons to full effect, propelling the Niners to five Superbowls.
Today, the storage world is seeing three advances that are comparable to Bill’s passing game: commodity flash, 10GbE and scale out architectures, which if leveraged properly in 2013 will change the enterprise datacenter.
This year, commodity flash has become truly compelling. Go check out Intel’s 910 series flash cards. The cards do 180K IOPS and cost $4K. SSDs like these are moving on to the PCIe bus exactly because they aren’t spinning disks. They simply do not belong in the SATA/SAS configurations where arrays that have been grinding it out on the ground for decades, yet this is exactly how array vendors are using them. Fancy bezels won’t change these fundamental limitations. Similarly, flash caches fronting these old school arrays bring about their own host of issues: namely that many storage workloads -- especially virtualized ones -- don’t benefit from naive, traditional caching.
The challenge with a PCIe SSD is that this raw performance needs to be paired with the right software stack, and that stack needs to be optimized specifically for flash. Flash can’t just be bolted on to enterprise storage features like caching, high availability, deduplication, and snapshots, the way these features are implemented needs to be rethought in the face of high performance flash.