Switch-level storage virtualization

Tech Closeup

Switch-level storage virtualization

Typically, storage virtualization has been performed through a host server or dedicated appliance. However, interest is building in virtualization at the switch level -- promising better performance and interoperability.

Storage virtualization creates a layer of abstraction that insulates an application from the hardware layer. This allows administrators to gather, organize, allocate and manage storage hardware without regard for the applications or servers that use it. That is, applications no longer need an awareness of particular disk volumes, arrays, NAS appliances or other storage locations. Data can easily be transferred and migrated between storage resources. Virtualization also permits storage resources to be altered and updated on the fly without disrupting application performance, generally reducing downtime. When properly implemented, storage virtualization eliminates forgotten or partially used disks, allowing superior storage utilization. This demands fewer drives, often translating into significant cost savings.

Switch-level virtualization promises to simplify maintenance and management of the infrastructure, but it's clear that the technology is still maturing. "With each iteration, the story is getting better and better," says Greg Schulz, founder and senior analyst at the StorageIO Group. "However, the total combined solution is still not robust enough yet for people to really take that leap of faith and run their enterprise applications on it yet -- outside of isolated cases, pilot programs, test examples or addressing particular pain points." In this special report, you'll find sections addressing the implications of storage virtualization at the switch, things to consider when implementing switch-based storage virtualization and the future of this technology.

Implications of storage virtualization at the switch

Storage virtualization is certainly not a new idea, and virtualization has been successfully accomplished for years. For example, VMware Inc. software has been an essential element of server-based virtualization, while systems like TagmaStore from Hitachi Data Systems Inc. (HDS) incorporate virtualization directly within the storage array itself. More recently, dedicated appliances, such as IBM's SAN Volume Controller, provide virtualization within the network fabric.

With such a comprehensive product repertoire to choose from already, the appeal of switch-based storage virtualization can seem unclear. Analysts concede that there is no new functionality at the switch -- users receive the same general set of features or capabilities realized with host, appliance and array-based virtualization. However, there are three compelling benefits that must be considered: simplicity, performance and interoperability.

Virtualization at the switch can greatly simplify the maintenance and management of the infrastructure. For example, server-based virtualization works well but can assume many different forms depending on the servers, operating systems and software mixes in operation. "You might have six different operating systems and four different platforms, and there are patches coming in for each version," says Arun Taneja, consulting analyst and founder of the Taneja Group. "By the third year of operation, you're just patching every day."

Moving virtualization to the switch not only eases maintenance but also unifies management. Instead of using a proliferation of management tools included with each storage box or server, storage administrators can move toward a common tool set to manage the virtual environment from the center -- the network. "You don't have to become a guru in Symmetrix and a guru in Shark and a guru in Hitachi TagmaStore," Taneja says.

Most traditional virtualization implementation schemes result in some performance penalty -- slowing the effective performance of the storage infrastructure. Placing storage virtualization at the switch can minimize any performance impact. "If it [storage virtualization] can be implemented in the switch with the lowest latency and least impact on performance and greatest transparency, then that's where it makes the most sense" Schulz says.

Switch-based storage virtualization also addresses the concern of interoperability -- ensuring that virtualization supports the platforms already running in a data center. In addition, the switch does not interfere with specialized functions, like remote replication. By centralizing virtualization at the switch, the virtualization engine becomes agnostic to platforms on both the server and storage sides of an enterprise network. "Network-based virtualization has that air of independence and can truly do a great job at heterogeneous virtualization." Taneja says.

A closer look at interoperability

Analysts all agree that a key selling point of switch-based storage virtualization is interoperability, but they disagree dramatically on the actual level of interoperability that users should expect from switch-based products today. For example, Schulz feels that interoperability remains a serious impediment. "The reality today is that when you look at some of these [switch-based] solutions, their interoperability isn't where it could or should be," he says, suggesting that products may need to be purchased with a close consideration of your specific environment.

Conversely, Taneja says that heterogeneous support is indeed a reality with today's switch-based virtualization products, pointing out that those independent third-party vendors all came to market with the sole purpose of creating a heterogeneous virtualization product. "They've done so, and they [products] work with all the major storage products in the industry," he says, citing the political clout of large vendors as the primary impediment to heterogeneous deployment. "It's [the vendors] not having the desire to let heterogeneity work."

Brian Garrett, analyst at the Enterprise Strategy Group, takes a conciliatory position somewhere in between, noting that interoperability should work today -- though it will take time to evaluate interoperability between vendors. "Solutions are not as interoperable as vendors would like to say," Garrett says. "Switch and software vendors are working through the interoperability matrix, and will take a matter of time to see broad matrices." Garrett points out that it took almost 10 years for Fibre Channel vendors to establish detailed interoperability matrices with disks, host bus adapters and other SAN components.

Ultimately, interoperability is clearly a major objective of switch-based storage virtualization, but the actual level of interoperability available with any given switch virtualization product today is a matter of debate. The best advice for users considering this technology is to discuss the issue with the product vendors involved and test interoperability in advance prior to any deployment.

Avoiding switch-based virtualization

Although switch-based storage virtualization may cause some service disruptions when initially deployed, there should be no noticeable impact on the storage infrastructure, operating systems, applications or network performance. Still, virtualizing at the switch may not be appropriate in several situations.

The investment in virtualization is best returned in large or complex environments where the technology can save dramatic time in storage maintenance and management. "We've seen customers reduce their storage costs by huge amounts," Garrett says. "The average savings being 20%-30%, but we've seen up to 80% savings in capital -- and good savings in human capital as well." Consequently, small or simple storage environments may not realize significant gains from virtualization.

Since heterogeneous interoperability is a key point for switch-based virtualization, homogeneous environments may receive limited value from this approach. "Some of the benefits that came with heterogeneous virtualization aren't there anymore," Taneja says. "But the value doesn't go back to zero."

Go to the next page of this article for implementation considerations and future directions

This was first published in June 2006