XtremIO Flash Storage Array

While attending VMworld this year in San Francisco, I had the opportunity to sit in on a briefing with EMC about its XtremIO flash storage array at Tech Field Day Extra. In my humble opinion, flash is going to be the way forward in the world of storage, especially as the cost of flash drives keeps coming down while their capacity keeps increasing. Will we get to a point where flash drives will be the standard for storage systems in the data center? Better yet, will spindled disks become the backup media stored off-site? Time will tell, but for now, flash arrays are quickly working their way toward becoming the industry standard for Tier 1 storage and beyond. 

EMC’s XtremIO flash storage array is a scale-out infrastructure based on a building block foundation utilizing EMC X-Bricks as its building blocks. It achieves between five and 120 terabytes, based on utilizing ten and twenty-terabyte-capacity X-Bricks, which can be clustered together to grow not only in capacity but also in performance. The X-Brick by itself is a storage appliance with a capacity of five, ten, or twenty terabytes. One thing that appears to make this solution stand out from standard physical flash drives is the in-line data reduction via deduplication and thin provisioning of the storage. The performance of the X-Brick scales out linearly, in that adding a second X-Brick doubles the available IOPS, and doubling the two X-Bricks to four X-Bricks means there are now four times the IOPS of a single X-Brick available. This linear increase in performance is due to the ability of the storage controllers in each X-Brick to access memory on other X-Brick controllers via RDMA internode communication.

The X-Brick RDMA controllers allow journalling of metadata to other X-Bricks in the stack. I have seen some other flash systems that utilize standard disk-based RAID algorithms in their data protection schemes, but XtremIO has developed a new, proprietary XtremIO Data Protection (XDP). For all practical purposes, this is something along the lines of RAID 6 in design. According to EMC specifications and marketing documentation, which give you the performance of a RAID 1 setup with greater capacity utilization of RAID 5, the XDP algorithm is used for the rebuilding of drives and the hot spare.

There is no inter-socket or inter-core sync of the system. XtremIO utilizes a content-addressing method for data location placement, using a segment data stream of data blocks with a size of a few kilobytes, and creating data finger prints of the data block to randomly disperse the data across the infrastructure. During the presentation about the data fingerprints, it was pointed out that statistically speaking, this is the most proficient way to achieve an “inherent balance” of the data to leverage all the resources evenly, based on the fingerprint value. Again, statistically speaking, I can understand how this algorithm for data placement offers an overall improvement. Nevertheless, this system of doing things still presents a minimal chance of errors. There could still be a chance: granted, a very small chance, but a chance nonetheless.

Another highlight worth mentioning is that the XtremIO has the ability to run at 100 percent capacity with everything  remaining balanced and a lack of endurance-related failures. After the deduplication process, the data will remain in an always-on in-line compression. You have the option to turn on data at rest encryption. All in all, this seems like one heck of a ride. Unsurprisingly, you cannot mix and match drive sizes in the system, and as for the price tag—well, that is one thing that is not for the faint of heart. I have always been told that you get what you pay for. I’ll leave it at that.