This week, at Pure Accelerate, the inaugural conference from the independent in the storage space which could be categorized as truly disrupting the industry, a NAS platform called FlashBlade was announced. Does it appear to you, as it does to me, a big game changer?
When Pure IPO’d recently, they’d had a block storage array based on all-flash which simply did what it was supposed to. It delivered fast fibre or iSCSI connected storage consistently, and with a highly scalable and very easy to manage interface. I have seen many examples of how implementations worked beautifully, even some in which the device was pushed into realms previously unanticipated where the platform simply delivered. In my opinion, the flash array that could simply worked.
One real hole existed in their product line, though, and that’s the ability to deliver NFS, file-based storage. I’ve had conversations with my customer base asking us to find a solution they liked as well as their Pure array, but that could deliver these services.
In comes the announcement of the FlashBlade.
The architecture is truly different than anything I’d seen in the storage space previously.
In terms of architecture, what we’ve got appears very logical, but truly game-changing from my perspective.
Each piece of storage in the chassis is an individual blade component, consisting of it’s own storage, processor and switch functionality. All these components are embedded on a narrow component on a slide which simply joins the aggregate once inserted.
They don’t stop there, though. These are not your ordinary components. The processor element is a shared workload between Intel and ARM (where there are two). Also, there’s a FPGA component to each blade. Architecturally, the workload are essentially divided up to their more appropriate processor in an effort to not only keep power utilization low at approximately 14watts, as well as forcing functions like Deduplication, Compression, garbage collection into their most appropriate parts.
Storage is accomplished not by Solid State Disc, but by NAND. The approach here was the removal of SSD overhead from the controller in the disc, as well as moving the storage closer to the processor, while removing overhead like SAS or SATA controllers, which only add to latencies in this configuration.
Each blade has its own switch, which then connect to the backplane of the chassis into which they’re residing. At that point, all the chassis then connect to a top-of-rack switch that ties all the individual chassis together.
Again, componentry is key. Everything has been designed in a modular fashion so that upgrades, and speed are the order of the day.
Currently, with a max per blade of 52TB, and a maximum of 1.6Pb per 15 blades, incorporating a rough estimate of 2.5:1 including compression and deduplication. When one assumes that the math on this is similar to a Raid6 config, available or usable storage likely conforms to approximately 1.1Pb per chassis. Each chassis is 4 Rack units in height, and thus, a usable capacity in 20 rack units of over 5 petabytes of usable storage. And this is today. The plans for the future in terms of sizing are only limited by the capacity and density of the NAND used. I can easily anticipate quadrupling this over the span of a year from launch. Seems really interesting to me. The press stipulates 16PB in one rack.