Original URL: http://www.theregister.co.uk/2012/08/20/emc_xtremio/
Let them eat cache: EMC unleashes ass-kicking flash box
Taking on Pure Storage, WhipTail and other all-flash array vendors
Soon EMC will have all the latest in storage tech: server flash cache, server shared-flash cache, and its very own all-flash array.
The acquired XtremIO technology  will be used to craft a networked all-flash array that is a platform in its own right – not just a cache like VFCache or Project Thunder.
EMC, like other storage and system suppliers, is using flash storage in servers as a cache layer between server applications and networked disk drive arrays to eliminate disk drive and network latency and thus speed up applications.
Its VFCache product is a PCIe flash card, OEM'd from either LSI or Micron, that provides read caching of data fetched from EMC VMAX or VNX arrays ... or any other array for that matter. Project Thunder is a collection of VFCache cards in a box that is shared between a number of servers and provides cache coherency. However, it is still a cache outlier of the back-end arrays and nominally part of the FAST tiering scheme – although it's not a cache and not a tier.
According to EMC's Chuck Hollis, VP for technical alliances, XtremIO will be a tier of storage in its own right. He recently blogged: 
What if you built a storage array entirely out of flash? And, in doing so, were able to free yourself from 20 years of legacy storage approaches as a result? That bright, shiny thought is what makes XtremIO's approach to block storage so compelling.
XtremIO's scale-out X-brick box has dual X86 controllers with 6 cores per processor, 24 cores in total, which run XIOS, the operating software. It deals with data in 4K blocks which are grouped to present block storage abstractions such as LUNs and volumes. Each block is hashed and XIOS provides global in-line deduplication as well as using the hash for addressing purposes.
X-bricks are scaled out using InfiniBand linking. Data is evenly spread across X-brocks and also across the flash components inside them. According to EMC, they are efficient in their use of flash capacity and also quite fast.
Hollis says: "The array delivers a predictable sub-millisecond I/O response time for every 4K block no matter what you happen to be doing: read, write, sequential, random, snaps, etc ... All data lives on flash. There's no need for data movement to adjust performance. All reads and writes are evenly distributed across available hardware resources. Each and every time. Snaps and primary volumes run equally fast." And the performance is fast even when the array is nearly full.
When EMC bought the company, the box had 4 x 8Gbit/s FC and 4 x 10Gbit/s iSCSI ports and redundant and hot-swap components. It delivers hundreds of thousands of random 4K IOPS and performance scales linearly as X-bricks are added to what seems to be a scale-out cluster. EMC engineers may tweak this hardware.
The XtremIO box is a silo – a very fast, thinly provisioned and deduplicated silo, but a silo none the less. Silos are okay though, when all you want is a silo. An obvious silo use case Hollis identifies is VDI.
Over time, Hopkinton may bring the XtremIO silo into the EMC fold by providing data management tools to integrate it with VMAX and VNX arrays, and even with Data Domain boxes. But this would require XTremIO deduped data to be rehydrated and, in the Data Domain case, deduped again. That's regrettable.
Performance-wise, this box will kick-ass with VMAX and VNX on block storage access. They have a capacity edge and an installed-base legacy edge – but if you want the fastest performance without needing back-end array integration, then the XtremIO box is better than either VMAX or VNX.
NetApp does not have an all-flash array in its product set and is concentrating on flash used as a cache in the array, the array controller and in servers.
This coming EMC box will compete with the dozen or so all-flash array vendors such as GreenBytes, Nimbus Data, Pure Storage, WhipTail and so on. It looks like it is going for the performance high ground and we'll see it possibly by the end of the year. ®