Josh Goldstein, Vice President of Marketing & Product Management at EMC XtremIO
Today EMC announced the general availability of the XtremIO (aka Project X). XtremIO fundamentally and irreversibly redefines what’s possible with all-flash array storage. Flash is supposed to be fast, but with XtremIO customers get much more. XtremIO irreversibly sets the bar higher in the flash array market. The XtremIO team has been working toward today since founding the company in 2009. Customers seem to agree, with already ~10PB of effective capacity purchased since April 2013 as part of the Directed Availability program.
Listen to Ehud Rokach, XtremIO’s General Manager and one of the original co-founders of the company and I discussing what it means for customers: http://www.youtube.com/watch?v=Ssdqo5VFF00.
There are four architectural pillars that are totally unique to XtremIO.
Content-Based Data Placement: XtremIO is a scale-out, multi-controller design, capable of delivering linear performance increases as capacity in the array grows. One of our fundamental design tenants was “inherent balance” so that the array leverages all resources evenly – all the time. In order to balance across a multitude of controllers and avoid hot spotting issues and back-end rebalancing processes, we developed a new way of determining where to place data by looking at the data itself with a sophisticated content fingerprinting process. This not only delivers better performance, better flash endurance, and makes the array easier to use, it has the secondary benefit of telling the array what it needs to know to deduplicate data inline in the data path at full speed. This type of algorithm would have been impossible to implement just a few years ago without today’s Remote Direct Memory Access cluster interconnect fabrics and it simply wouldn’t work with high performance on anything but an all-flash array.
Dual-Stage Metadata Engine: Any modern storage array virtualizes its internal addresses from the external addresses used by hosts. XtremIO took this one step further. We built a dual-stage internal metadata engine to virtualize the inside of the array. This lets the array maintain the validity of the external host to content-based address, while maintaining flexibility of the content-based address to physical array address. This freedom of data placement is a key enabler for XDP, our next technology pillar.
XtremIO Data Protection (XDP): Every other flash array in the market uses standard disk-based RAID algorithms. XtremIO leverages both the random access nature of flash and the unique XtremIO dual-stage metadata engine to deliver much lower capacity overhead, better data protection and much better flash endurance and performance compared to RAID. Perhaps more importantly, XDP is optimized for long-term operating conditions where overwriting existing data becomes dominant in the array. XDP allows XtremIO arrays to maintain their performance until they are completely full, thus driving the best economics out of flash.
In-Memory Metadata: Storage systems have always maintained a portion of their metadata in controller memory. This can lead to inconsistent and unpredictable performance. XtremIO delivers consistent and predictable performance because all metadata is stored in-memory. This makes XtremIO arrays impervious to changes in workload. It doesn’t matter what LUN sizes are used, whether there are random or sequential access patterns, or if there is locality of reference or not. But perhaps best of all, metadata heavy operations (such as inline deduplication, thin provisioning allocations and internal array copy operations) are conducted entirely in-memory—at lightning speed and without impacting I/O to the SSD portion of the array.
In working with XtremIO customers for over two years now, my favorite moments are watching their eyes get big as they realize what XtremIO can do. With XtremIO now available and at “EMC scale”, we anticipate those wide eyes and smiles to spread even faster.