Part 2 of 2
In part 1 of this two-part series, I discussed the evolution of All Flash Arrays (AFA) to meet ever-changing high-performance disk requirements. In this blog, we get into the various hardware that manufacturers are bringing to the market, including hardware that LPS Integration designs, installs and supports.
Violin was one of the first vendors to hit the market with a dedicated AFA in 2005. Whiptail (now owned by Cisco) also entered the market in 2009 with an AFA. It wasn’t until bigger, market-leading vendors like EMC and IBM developed or acquired an AFA offering that these systems disrupted the market and have exploded in popularity and desire. This is the technology that you absolutely need to have in your data center. These systems offer the same enterprise class technology as much larger storage enclosures such as battery protection in case of a power issue, RAID to support an unlikely disk failure (EFDs in the EMC XtremIO are estimated to function beyond 7 years without failure), and redundant connectivity into the disk subsystems.
At our company, we’ve become intimate with EMC’s AFA offering, the XtremIO, since it hit the market at the end of 2013. We also believe that for performance, reliability, data reduction services, redundancy and overall value for your money, the XtremIO is the best on the market. The XtremIO has a footprint of six rack units. In that small package, the XtremIO has the capability of sustaining 250,000 IOPs in what is known as a single X-Brick. The systems are linearly scalable, meaning that if you want 500,000 IOPS, you bolt on another X-Brick. Create a four X-Brick cluster and you get 1,000,000 IOPs. All of this performance happens in 22 rack units! For reference, it would take 5,556 spinning SAS disks to equal this amount of performance. And, EMC has provided the capability of growing an XtremIO cluster to eight X-Bricks this past summer, just in case you need to generate 2,000,000 IOPs in your organization. As a real-world example of the “space savings for performance” of the XtremIO, LPS Integration has recently replaced VMAXs with XtremIOs, reducing the floor footprint by six entire racks, while providing substantially more performance.
Each brick has a pair of storage controllers. While these may sound like storage processors, they are absolutely not. These two controllers contain large amounts of memory, spinning and solid-state disk, and are the true brains behind the performance operations. They literally take the SSD out of the performance plane. Incoming writes are put in memory, duplicated to the other storage controllers via a redundant Infiniband network, and then written to the disk once a full stripe can be written across all of the available disks in the disk enclosure.
The XtremIO, as are many of the other AFAs, is a thinly-provisioned enclosure, has deduplication and compression. What differentiates this enclosure from other AFAs is that deduplication and compression happen as data is being written. Other arrays on the market will do this after the data has been written or on a schedule, which reduces the life of the SSD by causing additional operations to occur. EMC has worked diligently to keep the amount of disk operations to a minimum. In addition to executing data-reduction services inline, they also found the only disk vendor at the time who did SSD garbage cleanup on the disk itself. This allows the storage controllers to do what they do best: provide incredible performance to the servers, and thus the end users. At the same time, the life of the SSD is extended. As for efficiency, depending upon the data pattern, the XtremIO can save you substantially on disk space. Virtual desktops deduplicate incredibly well (our test X-Brick gave us an overall efficiency number of 3500:1). Virtualized servers can deduplicate/compress to as much as 5:1, although generally the number is closer to 3:1.
So, what does all of this mean for your business? It means that by introducing a true AFA, vendors have brought to the market storage systems capable of sustaining your business’s disk workload in six rack units; which saves you money on infrastructure while exceeding your performance requirements. It also lets your storage architects and administrators get back to answering all of those storage requests without a concern in the world about where to put the workload.
Part 2 of 2.
Brian Ethington, Engineer