With a few tweaks to 3PAR, HP might crash all-flash array party
Blocks and Files Taking HP statements at face value, it would seem its flash array storage strategy is to extend its StorServe (3PAR InServ) disk-based storage arrays into the all-flash array world. This would mean taking on flash array start-ups like Pure Storage, SolidFire and Whiptail and facing down tech giant EMC - with its coming XtremIO product - and IBM, with its TMS RamSan line. Can this unique strategy work?
In October last year, HP said : "HP 3PAR is our strategic direction for solid state storage." At that time we commented "HP's strategy may involve extending the 3PAR array controller software to better manage a set of SSD resources." We now believe this to be pretty much a certainty.
The company already has an all-flash version of the 3PAR P10000 InServ array , which delivers P10000 model 800 performance in IOPS terms, based on using 1,900 300GB 15,000rpm disk drives, but using 512 solid state drives (SSDs) instead, and at 70 per cent less cost.
The 3PAR operating system is focused on storing block data on disk drives and does not contain facilities for storing data on SSDs unless the little beggars pretend to be disks. You can get much more performance, and endurance, from an all-flash array if you: a) treat blocks on flash as blocks on flash and not disk blocks and b) reduce the number of writes to delay the flash wear out point.
El Reg believes that HP is extending the StorServ OS to do this, to better use flash as a raw storage resource, also to be ready to use Memristor  storage - which may be ready in 2014.
Recent statements by HP's storage head, David Scott, support this idea. In a TechTarget interview  he said: "Today we have all-SSD solutions for StoreServ. In the future, we expect to seamlessly transition to all-flash and nonvolatile memory storage on 3PAR." The 3PAR ASIC uses a 16KB page or block size ands this means, Scott says, it's extensible to handling flash.
We're looking at a next-generation 3PAR ASIC that can handle solid state storage along with InServ OS extensions to do the same, with the aim of delivering comparable or better performance than start-up and/or stand-alone all-flash arrays. By having the StorServ OS extended to handle flash in its native state, the array uses will get all-flash array performance plus enterprise-class data protection and management facilities and, potentially, great connectivity to disk-based P10000 arrays for bulk data storage and data migration - with hot blocks sent to the all-flash product and cooler blocks (in data access rate terms) sent to the disk vault. HP's sub-LUN automated tiering could be used for this.
This is something that none of the stand-alone all-flash array products can do and is a beautiful strategy.
We expect HP to debut its flash-extended StorServ ASIC and OS some time this year. ®