Arthur Cole spoke with Shirish Jamthe, product development/product management executive, Virident Systems.
There is little question that data environments see considerable performance improvements following the deployment of solid-state storage. However, it is not simply a matter of replacing disks with flash cards and then watching your throughput jump through the roof. Proper utilization requires adjustments to both hardware and software resources, even down to the application level. Shirish Jamthe of PCIe flash specialist Virident Systems argues that effective due diligence is the only way to determine what type of flash is warranted, or whether it is needed at all.
Cole: When it comes to flash storage, the old saying, “There's a right way to do something, and there's a wrong way,” comes to mind. What would you say are the biggest mistakes enterprises make when provisioning flash storage?
Jamthe: The first mistakes come before it even gets to the provisioning of flash storage. There is a lot of buzz around flash and for good reason. It has the potential to provide unimaginable performance. However, first you need to determine if flash is right for your environment. You need to ask: Am I seeing poor application response times? Are I/O bottlenecks hurting my application performance? Am I seeing inefficiencies in my infrastructure to cope with performance requirements? Am I experiencing space and power consumption constraints in my data center? If the answer to any of these questions is yes, then it is time to move to the provisioning stage.
Once you begin provisioning there are three common mistakes that are made. Flash storage-friendly applications typically have a configuration setting that can be set to optimize the storage deployment. Unfortunately this is a step that is often overlooked and severely reduces the benefits flash was meant to provide. A second mistake is in the choice of the type of flash. Not all flash storage is designed to work with all applications. For example, with Microsoft applications it is important to leverage flash that supports Microsoft standard STORPORT architecture. If the applications are not carefully looked at prior to deployment there may be issues during deployment. The third, and biggest, mistake I see is being misled by peak performance numbers. While peak performance numbers can be impressive, it is only a snap shot in time and does not represent an accurate specification for long-term performance. The focus for the customer should be on sustained performance over time and across the various workloads you experience in your specific environment.
Cole: As a provider of PCIe-based flash solutions, do you think the protocol is ready to replace long-standing SAN and NAS infrastructure in the enterprise?
Jamthe: While I would love to think we’re at that point, no not yet. But we are close and it isn’t a shortcoming in the PCIe-based flash solutions that is preventing the move from happening. Most software applications were designed around the limitations of spinning disk and will take considerable effort to optimize for high performance provided by flash. Combine this with the fact that until recently many CTOs viewed deploying flash as a gamble with their career, there hasn’t been a lot of pressure to make applications flash-aware — and that is the hold-up.
I see this being a very different situation within three years. The introduction of Storage Class Memory (SCM) allows flash solutions to deliver memory-class performance with storage-class capacity and persistence. Concerns around reliability are no longer hindering the adoption of flash in the data center. As the price per GB gap between flash storage and spinning drives continues to narrow CTOs will no longer view deploying flash as a gamble and will realize the full capabilities of PCIe SSD. At that point we will begin to see PCIe SSD replace long-standing SAN and NAS infrastructure in the enterprise.
Cole: Do enterprises need to "future-proof" their flash infrastructure? What are some of the ways they can ensure flash will maintain its flexibility and availability in rapidly changing data environments?
Jamthe: With any technology deployment, it is necessary to think about future use cases before it is deployed in large scale. This is definitely true with the deployment of any storage and specifically with flash. A flash deployment in applications will give immediate benefits and ROI, but with the rapid growth of data capacity, needs will rapidly increase. This is especially true as users realize that they can store a lot more than before due to the increased availability of IOPS on flash. This leads to further consolidation and the need for more flash capacity per server. Choosing a vendor that provides a very high-density PCIe flash card that fits in all servers and scales in performance as more cards are added is key. You also want to select flash from vendors who understand your applications and their growth so you can ensure it will be flexible as your application needs grow.