Virtual Unity

Storage virtualization promises to smooth out capacity crunches.

Three years ago, Hunterdon Medical Center in Flemington, N.J., could claim one server for every two beds in its 176-bed facility. But for Alberto Cruz Natal, technical manager at the community hospital, that was nothing to be proud of.

"Our data center was overflowing with servers," says Cruz Natal. Worse, each of those servers had its own direct-attached SCSI storage device. When a server ran out of disk space, IT either had to buy another server or manually extend the server's storage partitions, a time-consuming and disruptive job. Meanwhile, some servers had excess capacity.

Reconfiguring storage devices wasn't exactly a chore that the IT group could afford to spend so much time on. The hospital was preparing to go live with a new clinical charting system, which precipitated a need to migrate the QuadraMed Corp. Affinity hospital information system (HIS) from its current platform—a 5-year-old Unix box from the former Digital Equipment Corp.—to a more powerful piece of hardware. The hospital had also just built a disaster recovery hot site 15 miles away, but it hadn't yet formalized a strategy to vault its data to that off-site location.

1pixclear.gif
1pixclear.gif

Simply Put

STORAGE VIRTUALIZATION makes heterogeneous storage arrays appear as a single logical pool of resources that can be managed from a single console, easing administrative tasks such as backup, archiving and recovery. However, some question the cost and potential for performance bottlenecks in some implementations, and nearly everyone agrees that vendors need to get better at integrating heterogeneous systems and simplifying deployment.

1pixclear.gif

Ineffective capacity utilization, growing data volumes, labor-intensive storage management, a need for better disaster recovery—this classic scenario is driving many users today to explore the world of storage virtualization. According to John Webster, founder of Data Mobility Group LLC in Nashua, N.H., data is growing at 60% to 80% or more per year for many companies, and storage administrators are spending 20% to 30% of their time on volume management tasks. It's clear that businesses need ways to simplify the job of managing all this data, and storage virtualization claims to help.

It does this by making physically separate and even heterogeneous storage arrays appear to be a single logical pool of storage resources, manageable from a central console. The goal is for data to freely flow among the various tiers and types of storage, depending on business needs, without disrupting the operating environment.

"The single most important attribute of any storage virtualization solution is the ability to mask complexity and thereby make manageable that which is increasingly unmanageable," Webster says.

Simplify, Please

There's nothing simple, however, about understanding all the different forms of virtualization on the market today and deciding which one is right for you. In Hunterdon Medical Center's case, the decision wasn't too difficult—its value-added reseller proposed that it move to a centralized storage architecture via a storage-area network (SAN). Because the reseller recommended moving the HIS system to an IBM RS/6000 (one at the hospital and one in the disaster recovery site), it also made sense to use IBM storage in the form of a high-end Shark array (one at the hospital and one off-site). IBM and DataCore Software Corp. in Fort Lauderdale, Fla., had an agreement to use SANsymphony software to virtualize the Shark array, so that was added to the environment as well.

Alberto Cruz Natal, technical manager at Hunterdon Medical Center
1pixclear.gif

Alberto Cruz Natal, technical manager at Hunterdon Medical Center

Image Credit: James Wasserman
1pixclear.gif

When this project was under way in 2003, storage heavy hitters IBM, Hitachi Data Systems Corp. and EMC Corp. weren't touting storage virtualization products. But now that they are—or, in EMC's case, are close to it—there are so many approaches to virtualization that it's difficult to decide what's best for your environment.

For instance, some vendors place virtualization capabilities on the storage-array controller itself (often referred to as array-based virtualization), meaning that you purchase both the storage and the virtualization capability together. Others place it on a server (often called appliance-based virtualization) that sits between the application server and the storage. Still others choose to put it on an intelligent switch (called network-based virtualization) that either takes an "in-band" approach, where the virtualization commands travel the same path as the data between the application server and the storage array, or an "out-of-band" approach, where the commands and the data take separate paths.

Even individual vendors offer a variety of approaches. For example, IBM's SAN Volume Controller (SVC) is an appliance-based system that's also available in a switch-based configuration. Meanwhile, its DS8000 is array-based, along the lines of Hitachi's Universal Storage Platform. EMC's forthcoming Invista is an out-of-band network-based product, which is the newest—and some say most promising—type of storage virtualization.

Although IBM's SVC currently leads the market with 1,200 installations, "the market is still very much in a state of flux," Webster says.

Up in the Air

No wonder most customers are still in evaluation mode with the technology. According to Tony Asaro, senior analyst at Enterprise Strategies Group Inc. in Milford, Mass., Hunterdon Medical Center is one of only 3,000 companies globally that have implemented storage virtualization today.

But Cruz Natal is pretty happy that he did. At first, he says, DataCore was "just another part of the system." Very quickly, however, it opened up a whole new world of possibilities. The most important, he says, is the ability to put any type of storage behind the DataCore virtualizer, including lower-end systems based on JBOD, FAStT and Serial ATA. This eliminated the need to keep non-mission-critical or less-accessed data on the high-end Shark system or manually move it to less expensive disk systems. "We now have the flexibility to buy different types of storage for different types of systems and manage it centrally through DataCore," he says.

Second, administration and maintenance are much less costly, Cruz Natal says. With a few clicks, administrators can create storage partitions up to 2TB for application servers. SANsymphony monitors the server's actual storage usage, enabling administrators to assign more storage to that disk pool on an as-needed basis. "We don't have to extend the partition or create a new one," he says. "We can just buy additional disk at a later point in time and assign it to the same pool."

And with a third DataCore server and a redundant RS/6000 server off-site, the hospital can also mirror data to the disaster recovery site, limiting downtime to a maximum of two hours.

Some observers say technologies such as DataCore's cause performance bottlenecks because of their location on the network. Bernard Shen, an independent contractor who specializes in storage architectures and server consolidation, argues that, given the cost of virtualization products and their performance hit, it can be just as effective in midsize environments to add more disk to the array rather than virtualize disparate arrays.

"In environments with medium to lower high-end capacities, I have not seen a true need to put a virtualization layer in place yet because of the cost and performance issues associated with it," Shen says. Costs include the price of the device, training people to use it and licensing fees. "If you have two SANs with 1TB each, you need to pay a license fee for 2TB," he says.

Shen also argues that adding a virtualization device adds a layer of complexity. "Vendors sell it as a single point of management, but that doesn't mean it's transparent to managers," he says. For instance, in environments where even the logical partitions are managed by the volume controller, you may not always know which disks are working with which servers. "You'd know a RAID set has failed, but you don't know which application is using that RAID without looking into it," Shen says. While Shen anticipates improvements in the technology nine to 12 months out, right now, he says, "I'm not sure that storage virtualization is necessarily universal for everybody."

Cruz Natal says the hospital doesn't experience performance slowdowns because it doesn't have a high volume of transactional data. "Bottom line," he continues, "it doesn't lock me into what kind of storage I use, which helps us keep costs in line, and it resolves the issue of training staff to manage the system manually, which lowers maintenance costs."

At the same time, Cruz Natal says he'd like management tools that give him dashboardlike visibility into things such as the status of partitioning volumes or the disk pool when slowdowns occur. "Better integration of all the tools becomes more critical because we have so many systems," he says.

Getting training staff to deal with this new architecture is crucial, he says. You need at least three people: one who's familiar with how the application servers interact with the SAN, one who understands the SAN fabric itself and an administrator who knows how to create new partitions, move volumes around and troubleshoot the virtualization server.

How to Choose

Array-based, network-based, appliance-based—the fact is, there's no "best" choice for virtualizing. It all depends on what you're looking for. "You have to look at what kinds of operations that the virtualization device, wherever it is, can offer you as a user and which are most important to you," Webster says.

The question is, what's your pain point? "Some people want to slow down their hardware spending, and others want to decrease their administration budget," Asaro says. Some might want to rearchitect their entire storage infrastructure, while others want to implement tiered storage in piecemeal fashion, he says.

Despite general agreement that virtualization adoption will take off in the next year and a half, everyone agrees that the vendors have to get better at integrating heterogeneous systems and simplifying deployment. Brad O'Neill, an analyst and consultant at Taneja Group Inc. in Hopkinton, Mass., compares it to the server virtualization world. "VMware has an easy-to-deploy solution with a lot of flexibility," he says. "It has to get to that level."

1pixclear.gif
Virtualization Type Perceived Benefits Trade-offs
Host-based

red_bullet.gif
Relatively low cost.

red_bullet.gif
Supports heterogeneous storage resources.

red_bullet.gif
Operating-system-dependent.

red_bullet.gif
Performance dependent on host processing resources.

Appliance-based

(Also called fabric-based)

red_bullet.gif
Supports heterogeneous storage resources and application servers.

red_bullet.gif
Future versions will be based on the ANSI’s Fabric Application Interface Standard (FAIS).

red_bullet.gif
Some implementations lack scalability and reliability for critical systems.

red_bullet.gif
FAIS still under development.

red_bullet.gif
Potentially adds I/O latency and opportunity for security breaches.

red_bullet.gif
Could introduce another layer of management complexity.
Array-based
red_bullet.gif
Can be optimized to underlying disk resources.

red_bullet.gif
Supports heterogeneous hosts but homogeneous storage resources.

red_bullet.gif
Proprietary implementation of virtualization.

Network-based

(Also called out-of-band switch-based or external full-function storage controller)

red_bullet.gif
Supports heterogeneous storage resources and application servers.

red_bullet.gif
Reduces complexity of switching infrastructure.
red_bullet.gif
Potential for proprietary implementation of virtualization.
NAS gateway
red_bullet.gif
Allows IP network to participate in storage virtualization.
red_bullet.gif
Support for back-end storage varies and is highly vendor-dependent.

SOURCE: Data Mobility Group LLC, Nashua, N.H.

Brandel is a Computerworld contributing writer in Newton, Mass. Contact her at marybrandel@verizon.net.

Special Report

Battling Complexity

Stories in this report:

Related:

Copyright © 2005 IDG Communications, Inc.

 
Shop Tech Products at Amazon