Flash storage has taken over the data center. From SAN to server, vendors have found creative ways to improve application performance using flash storage.
In the enterprise storage market, every major SAN provider now offers a flash option in traditional arrays to offer a Tier 0 level of storage. For instance, EMC has invested heavily in flash by purchasing and heavily marketing XtremIO; SolidFire, an enterprise storage startup, has leveraged all flash storage technology to challenge traditional storage vendors. However, enterprise storage isn't the only area in the data center to benefit from flash.
The increased manufacturing capacity has resulted in price drops and has made applications beyond enterprise storage a strong area for innovation. Software and hardware vendors have leveraged flash to increase performance throughout the data center. Let's take a look at the range of solutions available beyond network storage.
Levering SSDs with software
Flash-based innovations have made their way into servers. Putting solid-state drives (SSDs) into a server is a common practice to get an instant bump in performance without making any other change. The innovation has far exceeded just adding a pair of SSDs to standard drive bays to allow Windows to boot quicker. At the most basic level, hypervisors are starting to take advantage of SDDs to provide basic read cache; this has a significant impact on applications such as Virtual Desktop Infrastructure (VDI), which are read intensive.
Hypervisor vendors have followed the lead of PernixData, which is a leader in this new software category that leverages local SSD drives for cache to accelerate hypervisors. The vendor preaches ease of installation and management. An engineer only needs to install the software on an SSD-equipped hypervisor to reap the benefits of cache acceleration for read heavy virtual workloads such as VDI. PernixData is also doing interesting things with traditional RAM; its flagship product FVP enables a VMware cluster to create a RAM disk that can be used for cache acceleration for the entire cluster. Startup Infinio offers a solution that leverages server RAM to provide read cache as well.
Hardware vendors are also jumping on the flash bandwagon. Fusion-io makes PCIe flash acceleration cards that are designed to improve a wide range of applications. Unlike the software-only solutions, Fusion-io can accelerate individual applications such as SQL as well as hypervisor workloads. I've heard anecdotal testimonials from peers claiming that the performance doubled using PCIe flash acceleration for hypervisor workloads.
SanDisk and IBM have taken server-side flash to another level by delivering solid-state storage via the DDR3 data channel. It takes a little bit to conceptualize delivering block storage via the memory channel. SanDisk packs up to 400 GB of flash onto a DIMM to provide ultra-low latency block storage to local servers. By leveraging the memory channel, SanDisk and its initial server partner IBM can provide sub-7 usec (microsecond) latency via the DDR3 memory protocol. The UltraDIMM architecture has impressive maximums for a first-generation product.
IBM claims that a single 2U blade server can support up to (32) 400 GB DIMMs to deliver an impressive 12.8 TB of storage capable of delivering 4.5M IOPS. Just as in the enterprise storage and caching solutions, providing low latency access to local block storage has a tremendous impact on application capability. Latency-sensitive applications benefit most from this technology — think applications such as in-memory databases or high-frequency trading systems.
The bottom line
Flash isn't a silver bullet for every performance challenge. Prior to throwing flash at a problem, you need to understand your bottleneck. I/O may be the initial problem, but it can be caused by something other than the actual disk. For example, adding flash to a server with an undersized disk controller may not result in better performance and may result in decreased performance if there are more writes than reads. Similarly, adding flash to an array attached to an oversaturated SAN array will not result in better performance.
When the disk has been identified as the actual bottleneck, the industry has provided solutions to alleviate performance pains at the array or at the server.
- Fusion-io doubles capacity of its Flash storage cards to more than 6TB
- HP aims to take flash storage array costs below $2 per GB
- Google puts flash storage, HTTP load balancing to work in Compute Engine
- Violin Memory gives its flash storage an enterprise-friendly makeover
Disclaimer: TechRepublic and ZDNet are CBS Interactive properties.
Keith Townsend is a technology management consultant with more than 15 years of related experience designing, implementing, and managing data center technologies. His areas of expertise include virtualization, networking, and storage solutions for Fortune 500 organizations. He holds a BA in computing and a MS in information technology from DePaul University.