Why all the excitement (and investment) in flash storage?
There has been a good deal of investment in flash storage in recent months -- from acquisitions, to IPOs, to the funding of new private companies. The reason for this is simple. In the last decade, the tremendous growth of datacenter virtualization has caused an increase in storage performance bottlenecks. This, in turn, has hurt application performance, turning a key IT enabler into a potential business liability.
Flash is seen as a solution to this problem. As a result, we are seeing all of the incumbents in the storage and virtualization space announce initiatives around flash, and lots of activity from private companies focused on in this space. For example, we recently saw Violin Memory and Nimble Storage go public, Western Digital acquire Virident, Cisco acquire Whiptail, and the list goes on. Things are heating up, and this is just the beginning. Flash is the future of storage, and we have just scratched the surface.
Using flash in a virtual datacenter
There are two primary ways to use flash in a datacenter. It can be placed right in the storage array, and/or implemented inside the servers where applications reside. Because flash is extremely good at performance while not so good at capacity, some companies like Nimble Storage combine disk and flash together in their storage devices. The former is used for capacity, while the latter is used for performance. A number of incumbents, like NetApp and EMC, have done the same thing as this provides an easy migration path for their existing disk-based storage solutions.
Other companies, like Pure Storage and Violin Memory, are selling all flash storage arrays. They believe the performance gains achieved with flash outweigh the extra cost. Furthermore, they are banking on the fact that the price and capacity of flash will get better over time, obviating the need for spinning disks entirely.
Source: Fleshas via Wikimedia Commons.
The above strategies are certainly a step up over traditional disk based storage solutions. But, because they continue to tie storage performance with storage capacity, they can be costly to purchase, disruptive to deploy, and they may not solve application latency challenges as the storage device resides several hops away from the application itself.
This brings us to server-side flash. This is something that leading innovators, like Facebook and Google, have recognized as the key to solving their storage performance challenges, and have been leveraging in their datacenters for some time. It delivers maximum storage performance independent of storage capacity by leveraging cost-effective flash hardware that is placed right where the applications reside. In this respect, it solves the storage performance challenges of virtualized applications in a manner that is superior to flash in the storage.
But server-side flash is difficult to use across a virtual datacenter without disruptive changes to the applications, which has hampered widespread adoption of this technology. For it to become the de facto solution for storage performance, a new type of software is required.
Pioneered by my colleagues at PernixData, a Flash Hypervisor is software that virtualizes all server-side flash into a clustered tier that accelerates reads and writes to primary storage.
Flash Hypervisor software is installed inside the hypervisor kernel, requiring no changes to applications, hosts, or storage. Because of its clustering capabilities, it supports VM mobility and provides read and write acceleration with fault tolerance. This results in performance gains for all virtualized applications and storage performance that is independent of storage capacity.
The software allows companies to reap the benefits of server-side flash without the drawbacks. In this respect, it is reminiscent of the early days of VMware, where a disruptive technology that is non-disruptive to deploy changed the face of computing forever.
Flash is the key to solving storage performance challenges in virtual datacenters. It can be implemented in the traditional manner, whereby shared arrays are beefed up (or replaced) using new flash drives. Or, it can be implemented in servers in conjunction with software to decouple storage performance from storage capacity. Far from a "flash in the pan," flash in the datacenter is here to stay, and it will be at the forefront of storage innovation for the foreseeable future.
ó Poojan Kumar is co-founder and CEO of PernixData.