Optimized Flash Storage Platforms: Disrupting the Economics of the Data Center

Erik Ottem is Director of Product Marketing for Violin Memory.

Imagine that someone devises a solution that reduces your hour-long commute to a couple of minutes. That’s the equivalent of what optimized flash storage platforms do in the data center.

Most servers are I/O constrained so speeding up storage can boost a server’s output considerably. Compared to a 60 hard-drive RAID box delivering 9000 IOPS in total, the optimized flash platform delivers 1 million IOPS at blindingly fast speeds.

This extra horsepower can really impact your data center cost profile. Like many data centers, you probably have “farms” of the fastest hard drives you can find to get enough IOPS for your servers. Capacity isn’t the issue; and, in fact, most storage is significantly over-provisioned to provide adequate performance. Hard drives, for a decade or so, have maxed out at around 300 IOPS. Putting an optimized flash platform into your SAN will supercharge it, provide the needed IOPS to load up your servers, and allow those hard drives to be relegated to tertiary storage, where their lethargic performance no longer matters.

The half-step to an optimized flash platform is the hybrid array. It uses flash as a cache, with auto-tiering software moving data out to the secondary hard drive storage. The problem with hybrid arrays is that a cache miss, where the data must come from the HDD, creates an I/O that may be 100 times slower than data pulled from flash. Tiering may be rational if there is significant money to be saved, but with the new economics of flash, tiers for active data can no longer be justified.

The major implication of this performance is workload consolidation. Instead of buying 100 new servers to keep up with your workload, an optimized flash platform makes your existing servers considerably more efficient. Workload consolidation is made possible through the combination of virtualization and flash storage. But not all flash is created equal. SSD-based designs have latency spikes as the arrays go through garbage collection. A flash-optimized design doesn’t use SSDs, but manages flash across the entire array as a pool of flash for consistent low latency. Consolidation with the right platform provides a great return on investment.

More efficient servers mean fewer servers are required since they spend far less time waiting on I/O, so CPU utilization can hit 90 percent or more. In addition to less server hardware, that’s 100 sets of software licenses that you just avoided. It also means less labor managing the applications and operating system updates and savings on power, cooling, hardware maintenance and support.

System users stand to benefit as well. Consistent low-latency storage impacts job run times. A task that took three hours might be completed in minutes. When compiling code, run times drop from hours to minutes, and the productivity of your expensive software developers goes way up. Low latency also opens up new business opportunities at the corporate level. For exame, one major European telco found that its billing software couldn’t keep up with users and employed an optimized flash platform and added $140 million to its bottom line.

The bottom line is if you haven't tried a flash optimized platform, you should. They drop easily into existing SANs, and the performance benefits and enterprise data services will be obvious in a few hours of use. When the total cost of ownership is considered, an optimized flash platform is actually less expensive than legacy disk-based storage, hybrid storage or even SSD-based storage.

IDC predicts that all-flash array revenues will top $41 billion in 2016, so it just may be time to declare disk dead for active data.