Vishal Misra is founder and chief scientist of Infinio.
Does your data center have flash fever? If it doesn’t, I’m sure some of your neighbors do.
Declining prices and a rich selection of flash-based products are fueling an adoption of flash storage across the enterprise. However, it still comes at a price premium that many organizations can’t afford, leaving ample space for alternatives that offer some of the same benefits at a fraction of the cost.
Many of today’s servers are loaded with abundant, extremely fast memory and processing power that often sits idle and can be exploited to provide flash-like performance. Flash-based solutions may seem like silver bullets that can magically fix every issue in your IT system, but despite the benefits they can add, they can also introduce new complexities and costs that can easily be avoided.
If you’ve already adopted flash, don’t make the mistake of discounting its shortcomings rather than seeking out alternative options that might be a better fit for your environment. To harness the full potential of storage, consider the performance and business issues you’re looking to solve in the first place. You may find that modern architectures exist that can deliver flash’s results at a fraction of its cost.
Diagnosing Your Data Center
If you’re using flash and unsure if it’s living up to your initial expectations, you’re not alone. For most organizations, storage issues come down to cost, performance and capacity. Look out for these major symptoms of flash fever:
- You’re paying too much for what you need. Flash storage guarantees consistent performance for highly transactional, read-intensive workloads. On the other hand, the high I/O requirements of virtualized environments, such as server and desktop infrastructures, are a great fit for DRAM-based server-side caching technologies that are not dependent on flash. Performance requirements are often in milliseconds, but flash provides microseconds, which is not needed and is an unnecessary expense that can and should be avoided.
- Your applications are write-intensive. Flash devices have a finite number of program-and-erase cycles, meaning a high number of writes can make the platform economically unrealistic and impractical. In this case, the fixed lifetime of solid-state devices (SSDs) accelerates the refresh cycle and adds another recurring expense.
- You need high storage capacities, but they have to fit your budget. It’s true that design constraints prevent SSDs from reaching the same capacities as hard disk drives (HDDs). Many of today’s SSDs are below 1 terabyte (TB) per unit, compared to 10TB provided by some of the leading-edge HDDs, averaging $.03/GB. SSDs will most likely never provide comparable massive capacities, and their price per gigabyte (GB) can measure up to 20 times higher than typical HDDs. Capacity-optimized storage arrays combined with server-side storage acceleration technologies can deliver the most dramatic economic benefits.
Making It Work
For some IT managers, one way to recover from flash fever is to adopt a host-accelerated model for storage performance. These systems use scale-out platforms to decouple I/O performance from storage capacity and provide the best performance at the lowest price. As a result, users can access server-side, scale-out performance that can be paired with any capacity architecture, including legacy arrays. Some users choose to realize immediate performance benefits on an existing storage area network (SAN) or network-attached storage (NAS) array.
However, the most dramatic economic benefits can be achieved when this architecture is a part of a new deployment, combining storage acceleration software on the host with a dense, capacity-optimized array for ultimate flexibility, control and efficiency.
For issues rooted in storage capacity, shingled magnetic recording (SMR) drives are emerging as the next ultra-dense storage medium. The newest 10TB SMR disks are priced at just 3 cents per gigabyte, making them ideal building blocks for inexpensive capacity. On the performance side, non-volatile, dual in-line memory modules (NVDIMMs) are emerging. NVDIMMs plug directly into the memory bus or DIMM slot on a motherboard instead of the higher latency PCIe. Writes will be completed in fewer than 10 microseconds, an order of magnitude faster than the current mainstream flash can handle.
What ever way you choose to recover from flash fever, the potential of existing arrays and new storage deployments will grow when you simply consider the other options in today’s IT landscape. At all costs, however, you should avoid wasting any resources – memory, processing, budget or otherwise.
Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating. View previously published Industry Perspectives in our Knowledge Library.
