Over the last decade, virtualization has revolutionized the data center by more effectively leveraging the CPU and memory resources on the server to increase hardware utilization and reduce operating expenses. Of course, as with all things performance-related, once a log jam is alleviated at one choke point, the bottleneck shifts to another part of the system. In the current environment, that limiting factor is storage performance.

The historical method of adding disk spindles to increase performance has proven to be expensive and inefficient. Speedy flash technology has recently emerged as the leading candidate to answer the storage performance challenge. But how is flash best incorporated into the infrastructure?

The obvious choice is to add it to the shared storage system. This can be done by upgrading an existing array or replacing the legacy system with a new hybrid or all-flash platform.

A recent PernixData, Inc. survey of 380 IT professionals indicates that many are struggling with this issue. Half have either implemented or plan to implement a hybrid storage solution while 28 percent already have or will deploy an all-flash array. The advantage of introducing flash into a virtualized infrastructure is that performance gains benefit all the hypervisor hosts uniformly. However, the primary drawback is that one of the leading inhibitors of storage performance is latency in the storage area network (SAN). Speeding up the storage system does nothing to reduce the time needed for data to traverse the physical distance between servers and storage. This approach also creates a pool of IOPS that can’t be allocated to individual virtual machines (VMs), which limits flexibility. Finally, this tactic ties storage performance to capacity, which leads to wasted investment in overprovisioned storage arrays.

Another option is to include flash in the server. This results in the greatest performance gains because it brings the fast storage medium closer to the VM and therefore the application. It’s also a more cost-effective way to incorporate the latest in flash devices as the rate of technology refresh occurs more rapidly on the server. That’s why 44 percent of respondents to the survey have either implemented or plan to implement server-side flash.

The downside, though, is that acceleration is localized to the physical host, which creates erratic performance. The VMs that happen to be on the box with flash storage run much faster than their network peers that lack flash. Also, if those VMs are live migrated (for examples, with VMware’s vMotion feature) to one of the other hosts, the application may suddenly experience a dramatic slowdown.

A new model has recently emerged to bridge the gap between these two modes. Hyper-converged systems essentially combine the server and storage systems into a single unit. By collapsing the infrastructure topology, the storage network latency penalty is minimized and flash is better used. These appliances also generally include a clustered file system that enables more predictable performance scaling.

The problem with the hyper-converged model is in implementation. By definition, it’s incompatible with the existing server and storage infrastructure and therefore requires a full replacement of both, which is disruptive. Once this “forklift” upgrade takes place, it may introduce vendor lock-in and limit customer flexibility.

New virtualization software solutions are addressing this gap. They create a clustered pool of flash resources across hosts that is used to accelerate reads and writes to primary storage. This ensures the fastest VM performance without requiring any changes to existing storage or hosts. In addition, they ensure seamless VM mobility across hosts. As a bonus, they can also be used for fault tolerance within the server tier by replicating between hosts.

When virtualization software is implemented with server flash, storage performance is decoupled from capacity for optimal VM performance and true scale-out growth. Not surprisingly, 42.4 percent of the IT managers polled have already evaluated such software, and a whopping 70 percent said they were likely to purchase it in the near future.

If flash is becoming more pervasive in your data center, you need to optimize its placement and maximize its impact. Learn about the pros and cons of each approach so you can do what’s right for your organization.

 .

About the Author

Jeff Aaron serves as PernixData’s Vice President of marketing, where he is responsible for global promotion of the company’s storage acceleration software. Prior to PernixData he was VP of marketing at Silver Peak Systems, a leading provider of data acceleration software.

photo credit: zachstern via photopin cc