Fabric-based intelligence: but do a reality check on the switches before you buy.

Switch vendors are actively adding more intelligence at the fabric level. What that intelligence should be is open to debate (and different vendors are promoting different types) but most agree that managing the SAN through the infrastructure is very promising. Development is generally centering on copy and move functions, such as replication and remote mirroring; virtualization, provisioning and storage management; and internet-working and utility-based models. With so much active development going on, both established switch makers and starry-eyed start-ups are preaching the benefits of concentrating storage management and protection in the infrastructure. The question is: Is this a good thing?

Rob Emsley, HP's product marketing manager, Network Storage Solutions division, thinks there's not really one right answer to that. "Clearly, within the storage area networking industry, the fabric is a core component of the environment. It's like the hub in the wheel--everything is connected with it. The power of the network is directly proportional to the number of elements connected to the network." The value of the SAN increases according to the number of connections. That's why fabric-based intelligence gets so much focus: people recognize that network management and the capabilities of the network are key components to its value. Since Fibre Channel SANs are more expensive and complex than IP networks, SAN buyers are looking for ways of getting maximum value out of the investment. Adding capabilities to the environment multiplies the value of the ROI dollars.

Sounds good, and it can be--but the corollary to concentrating intelligence in the fabric is the absolute need to protect that fabric. Once you start using switches for more and more management functions, administrators must make sure they failover and replicate those switches, as well. If there's a bank of ports or a large director switch that's handling storage resource management, archiving and remote mirroring, those pieces had better be fully redundant with no single points of failure. Given price, complexity and risk levels, many vendors--including established switch vendors--are still leaving framework management to software. For example, Brocade puts robust security functionality into its switches, but leaves management to storage management software.

Still, most switch makers are charging ahead with intelligence development. Mark Stratton, McData's director of Solutions and Alliances programs, describes the switch vendor's viewpoint. "We're looking at managing more of what's in the fabric, also providing more information about relationships of what we're managing. We'll continue to be as aggressive as we can in this space." Switch intelligence vendors are primarily concentrating on several different developments, including copying, storage management, and utility storage computing.

Copy and move: These functions include replication, remote mirroring, and online backup. They also control tiered storage, which migrates data to different storage targets depending on characteristics.

Storage management: Virtualization is a big development center for switch makers, since virtualization significantly improves volume management and provisioning between different storage arrays.

Utility storage computing: The utility model sees the network infrastructure as a dynamic storage manager instead of a costly investment.

Copying

Fibre Channel fabric is well suited for copy and move functions, the foundation of backup and replication, These functions are largely proprietary today, and dependent on different vendors' backup and replication applications. By putting the copying mechanism in the fabric rather than relying on the sub-system host, it is possible to have a heterogeneous, open-based connectivity approach to replicate, mirror, and backup between arrays on local and remote sites. The fabric may control such functions as replication, remote mirroring, and tiered storage schemes.

Tiered storage operations migrate data to different storage targets based on policies. For example, a production database would retain its active data on high-end arrays, while a tiered storage manager would archive inactive data onto less expensive ATA disk. Given data parameters, some of the data would remain on ATA indefinitely, while other data would in turn migrate to online tape libraries or off-line vault storage. A number of software packages do this today, but some switch intelligence developers believe that the fabric would handle tiered storage much more efficiently in open environments.

Storage Management

Storage administrators want and need to automate repetitive tasks with storage management tools. Software already does this, but switch makers believe that the fabric is better suited to managing SAN-wide storage than host-based applications. For example, virtualization is the foundation for automated provisioning. Some switch developers prefer to see virtualization in the physical data path, where it can dynamically allocate storage to a pool of shared resources. Virtualization-capable switches might contain a policy engine that manages layers of sub-policies, which in turn automate provisioning between the arrays. Stratton added, "We're looking at managing more of what's in the fabric, also providing more information about relationships of what we're managing. We'll continue to be as aggressive as we can in this space."

Harish Nayak, co-founder of Maranti and vice president of marketing, said, "Look at what intelligence means, virtualization is not the end-game. IT is a tool to achieve provisioning of required capacity to applications. This capacity has certain characteristics like performance, availability and security." Intelligent network infrastructure may be able to provision and manage storage by application needs--altering operations based on data characteristics like time and date, users, and application service levels.

Maranti's Rick Walsworth, director of product marketing said, "Customers want the ability to have a dynamic pool of resources to move around based on time of day, seasons of the year, whatever. Otherwise, they have to roll in a new array every time their demand goes up." Dynamic allocation is vital to efficient provisioning: many network administrators only provision to thirty percent because they're anxious to maintain their applications' service levels.

Utility Storage

As SANs consolidate and grow more complex, they may need to evolve from an efficient way to share libraries and arrays, to a utility that can efficiently utilize and control data according to application demands. For example, in a utility model a switch would house a policy engine containing bandwidth requirements for each connection it controls. The switch would grow or shrink the bandwidth according to the policy that governs the connection's data traffic. One advantage of basing bandwidth allocation in the switch is that it can be very granular (for example, allocating bandwidth according to time of day, or during fiscal end-of-year for accounting systems).

Eric Blonda, director of product marketing at Sandial, believes that "it just comes back to control, that's what customers are looking for. They want a smart infrastructure they can set up in the utility, and then control how that bandwidth resource gets appropriated based on the task on hand. And that changes."

Internetworking is another Next Big Thing. Right now, Fibre Channel is similar to a flat Layer 2 network with no hierarchy or subnets. Since subnets would help to connect remote SANs and manage large consolidated SANs, some switch vendors are working to develop a subnet-capable fabric. Some virtual subnets like Cisco's proprietary VSAN have already hit the market, and the iFCP protocol gives Fibre Channel subnet-like features. But to make internetworking successful across the storage market, it should be a fabric-based technology with standardized techniques. Storage networking standards such as SMI-S will be crucial for this type of open internetworking development.

Get Some Backbone

Some switch vendors refer to large-scale intelligent switches as backbone switches and treat them as natural extensions of director-class. Michael Welts, executive vice president of marketing at Sandial, said, "As the model of SAN begins to get exploited, the need for extending that director-class market into a greater capacity platform, a backbone platform, seems to be the next natural progression in the high-end solution. You're ending up with an edge, a core and a backbone." Not every switch vendor accepts this terminology, but all of them are beefing up their high-end products to contain more and more levels of intelligence.

The startups developing backbone switches are investing them with heavy-duty storage management capabilities. These backbones are meant to support consolidated enterprise SANs with huge port densities and large-scale scalability needs. Some of the new switches will work only on their own fabrics; others will work with existing switches from other vendors. Blonda added, "The network itself is taking on more of a hierarchical position. We're seeing a backbone take shape where there might be more extensive management capabilities and performance capabilities at the corn of the network, while the edge of the network is used as an aggregation point, where as many resources as you can will be SAN-attached."

Whatever an intelligent switch may offer, buyers must be very sure that adding any intelligence to their switches isn't going to threaten their network. That means that innovative switches from start-ups, especially the ones that won't work with multi-vendor switches, will be a hard sell. Still, most SAN developers and customers know the problems with managing and provisioning large consolidated SANs, and switch-based intelligence might have a strategic approach it was lacking before. The tactical issue has always been to rationalize costs around infrastructure that has been overbuilt over the last couple of years. The strategic issue is moving away from a necessary-evil view of infrastructure, to an efficient utility method of delivering services.

Commenting on the move from tactical to strategic development in the industry, Mark Davis, senior vice president of marketing mad business development at Candera, said, "SANs today are in a place where the entire infrastructure is falling under its own weight. While I think SANs are wonderful things, I also think we've hit the point now where we've got to go to the next level. We've got to put intelligence in the network. That's the big mega-trend that's really driving development."