What's New

22nd Oct 2008

What to Ask Before Launching a Storage Virtualization Project

Industry watchers offer five key questions.

NASA's Infrared Processing and Analysis Center wasn't shooting for the stars when it turned to virtualization to meet its storage needs. IPAC's cash-strapped effort to record images of our universe -- up to 30 million objects captured each night and 42 billion records over the life of the project -- required big storage capabilities, and the engineers needed them fast and at a low cost.

"We were trying to find a way to step outside of the normal storage purchases to meet our 'high performance and high availability on a budget' requirements," explains Eugean Hacopians, senior systems engineer at IPAC. IPAC had already purchased a shared-storage system from Seanodes, to get control of its clusters with multiple compute nodes.

But Hacopians soon learned that he could put storage on the nodes and that they could work as compute servers and storage servers -- without additional costs or upgrades.

"In general, I'm not really fond of virtualizing things," he says. In his mind, "everything has its own place. But it's a solution that fits a need."

Indeed, virtualization can offer a solution for many storage challenges. But it can also be costly to buy and complex to implement, and it might require you to purchase equipment you didn't need before, such as new switches or servers. How do you decide on the right approach and choose the right vendor? Industry watchers suggest five key questions to ask yourself and your prospective vendors before selecting a storage virtualization technology.

1. What problem are you trying to solve?
The term storage virtualization has become a catchall phrase used to refer to many types of technology that make more efficient use of your storage assets. It can also bring these assets under a single management umbrella with a single point of control.

Since storage virtualization comes in all shapes and sizes, first determine what level of storage you're trying to optimize. Is the pain point at the block level, file level or tape library?
For block-level storage, virtualization can help consolidate large, disparate soft assets in the form of storage tiers, or it can simply bring them all under one roof.

At the file level, virtualization comes in handy when companies develop too many islands of network-attached storage. "[If] your users are storing data all over the place, and you can't back them all up under a single roof, you use storage virtualization to bring all that under a single umbrella, and everyone accesses it through a common [naming convention]," explains Ashish Nadkarni, principal consultant at GlassHouse Technologies.

At the tape library level, virtualization is used for making online storage appear as tape to the backup software.

2. Do you want host-, network- or array-based virtualization?
When deciding what type of virtualization is best, "it really comes down to what problem you're trying to solve and what kind of vendor affinity you have," Nadkarni says.

For most IT units, having host-based virtualization is a given, since volume managers run on the host. More often than not, you'll see host-based virtualization in a storage-area network (SAN) environment.

"Array-based virtualization is more of a function of which vendor you're going with for your primary storage," Nadkarni says. For example, with some Hitachi Data Systems storage products, virtualization can be deployed by enabling an existing software key within HDS's Universal Storage Platform or its Network Storage Controller. "So you'll go with array-based if you plan to buy a Hitachi frame for your Tier 1 storage," he says.

Network-based virtualization is typically used if you plan to make your SAN a multiprotocol storage network and in doing so are porting the network intelligence -- which also includes virtualization. Cisco Systems and Brocade Communications Systems are now coming out with products that let you virtualize within the network, without the need for an array or a host to do anything.

Some products blur the lines between host-, array- and network-based virtualization. "Products like Seanodes' would be considered host-based virtualization because you're virtualizing over the nodes," Hacopians explains. "You could also think of it as network-based, because you're virtualizing and spreading it across and letting the network take care of itself."

Industry watchers agree that virtualization might be easier to implement and cost less if IT groups stick with their vendors. "If you're primarily in a Hitachi environment, for example, array-based virtualization is probably going to make the most sense," Nadkarni says. "If you're a Cisco SAN, and you already have the infrastructure to implement Cisco virtualization, then network-based makes more sense."

3. How much complexity can you handle?
Host- and array-based virtualization are usually the easiest to implement, experts say. Network-based systems are often the trickiest because there is no direct way of virtualizing in a network. Most IT shops use third-party appliances. Cisco's system usually requires users to buy enabler software or an appliance or other third-party tool that sits alongside it, Nadkarni says.

"Then you have to figure out whether it's going to be asymmetric or symmetric," he says. "Where are you going to store your depository? What services do you want to provide? What arrays are you going to virtualize?"

In an array-based setup, "you take your second-tier arrays and just virtualize them behind your existing arrays. It's one view to the whole world -- like having one entrance to the office," Nadkarni adds.

Gene Ruth, an analyst at Burton Group, says the simplest approach is to choose an all-inclusive system, add appliances and then link them. But beware of diminishing returns. "At some point, it just gets complicated, and it may not be worth it when you aggregate too many appliances," Ruth says. "Then you have to ask yourself, is it better or are you getting this lowest common denominator?"

The hardest part is the planning phase, says Roman Perez, systems engineer at Business Technology Partners. "If you have a big company with thousands of servers, you have to do it little by little, and that's a big project," he says.

4. What's your budget?
"It's always cheaper when it's well thought-out and part of a bigger project," Nadkarni says.

Your budget will depend on the type of virtualization you need. Block-level virtualization is cheaper if you implement it as part of your upgrade. If you're buying or implementing a new SAN, then incorporating storage virtualization within the SAN is more prudent than buying off the shelf. "It tends to be pricey because you're now trying to 'a la carte' it. Do it as part of a larger upgrade so you can bundle some costs into the upgrade itself," says Ruth.

He also recommends that IT managers compile a spreadsheet to compare those scenarios. Adding appliances could be cheaper in tough economic times, but be sure to amortize the cost over three to four years, since older systems will eventually need to be replaced.

For virtual tape libraries, it's important to correctly estimate virtualization needs -- or risk buying much more capacity than you need.

5. Do you have an exit strategy?
Keep in mind that network-based virtualization can get tricky, Nadkarni says. "It can get a little complex over time, so you have to make sure that whatever architecture you're implementing [can be withdrawn from]," he says. "You shouldn't be stuck with it."

Most storage virtualization products create metadata from your data. That's how the storage objects they virtualize are managed. "Unvirtualizing means figuring out how to reappoint your metadata back to original data," Nadkarni explains. "The second problem is, your data could be across multiple storage areas or multiple objects. In that case, you now have a challenge of trying to present the same data, in a committed manner, back to the host again."

Storage virtualization isn't an insurance policy against sloppy practices, Nadkarni says. "It's almost like taking a dirty room and stuffing everything that's out of place into a closet," he says. "You really need to put things back into their place. So storage tiering or other good storage practices need to be taken care of first. Then you can move to the next step and implement storage virtualization."