Using SANs for business continuity

Be wary of looming interoperability issues.

By Betsy Yocum
Network World, 05/06/02

Business continuance - the ability to get your network up and running after a disaster with minimal impact on business - has emerged as the leading driver behind storage-area network and network-attached storage adoption in the post-Sept. 11 world.

"People now realize that SANs are a key to gaining disaster tolerance," says Ron Totah, technical marketing manager for Brocade Communications Systems. Until Sept.11, network managers didn't envision the magnitude of the catastrophe that many businesses in New York faced. Totah says managers are not only planning to duplicate networks, but also are tripling them in case the original resources are destroyed.

Businesses that use a SAN to mirror data in real time over optical connections can recover much faster than those that are still sending reels of data and performing hourly synchronization processes. SANs that support longer distances offer an advantage by letting companies store data farther away. Brocade's SilkWorm switch, which we tested (see review) now supports distances up to 74 miles.

But these extended storage networks will need to encompass a range of technologies including Fibre Channel, Internet SCSI (iSCSI) and 10G bit/sec Ethernet, and will have to accommodate SAN and NAS technologies in tandem, which requires much more interoperability than the storage industry now allows.

In fact, while the SAN and NAS markets have been targeted for significant growth over the next five years, our research shows there are a number of technical, interoperability and standards-based stumbling blocks in the way of widespread adoption of these storage technologies.

Fibre Channel, an ANSI-based standard that provides a high-speed, short-distance connection, is likely to stay in the data center, according to Doug Ingraham, Cisco's senior manager of product strategy and management.

Brocade's switch earns big numbers
Interactive Buyer's Guide (Excel spreadsheet)

But in lower-end server environments where users haven't embraced Fibre Channel mainly because of the price tag, iSCSI will have its biggest audience. ISCSI, a standard currently being finalized by the IETF, defines how SCSI commands traditionally used for local data communication between a host's CPU and attached peripherals are sent over an IP connection.

Dale Lafferty, vice president of marketing for Inrange, says Fibre Channel is the protocol of choice in storage networks because it has a proven track record in terms of performance and reliability. Inrange's flagship directory switch, the FC/9000, 64- and 256-port, is based on Fibre Channel.

Fibre Channel's incumbency - there were 263,441 Fibre Channel director-class switch ports installed worldwide in 2001, according to IDC - is indeed an advantage in the market, says Marc Staimer, president of Dragon Slayer Consulting. Fibre Channel has undergone huge amounts of interoperability testing among industry groups with good success and has established a beachhead in corporations because of its low latency and high-speed performance. But it's still expensive compared with Ethernet and requires a higher level of technical expertise to implement and manage, Staimer says.

ISCSI, on the other hand, paves the way for managing storage networks over long distances, linking "islands" of SANs into an integrated entity at a lower price. But the struggle to gain an industry standard for iSCSI could prove the major stumbling block to its adoption in the short term. The IETF has yet to ratify the draft standard for iSCSI. But according to Cisco's Ingraham, major points in the draft have already been agreed on, and vendors currently are working on proving interoperability based on those points (See iLabs iSCSI testing, page 66).

On the issue of SANs and NAS convergence, Staimer points to the fact that vendors such as EMC, IBM and NetApps already are shipping products that can operate side by side. Convergence currently has two faces. The first comprises a NAS head (or purpose-built server) that sits on a SAN in front of block storage. This server, typically a Windows NT/2000 server with an host-bus-adapter fiber network interface card, provides file-level management for the SAN's block-level storage. The second face is a NAS system that has integral SAN connections, eliminating the need for the extra server mentioned previously.

Staimer notes that the storage characteristics of SAN and NAS must be addressed in terms of users' applications. Block storage is used when fast performance is important, while NAS-based file storage is best when ease of use is an issue. So while SANs, which support block storage, and NAS, which supports file storage, are very different, each has a role to play in storage networks.

2G bit/sec and beyond

In terms of the speed necessary to support growing enterprise storage needs, our sources agreed that 2G bit/sec Fibre Channel is gaining its legs, beginning with its debut on director-class switches (see review). Although 10G Ethernet will play a role in the future of storage networks, that's not likely to be viable before 2003.

McData CEO and Chairman Jack McDonnell says that pushing Ethernet full 10G bit/sec bandwidth is not even close to reality because IP stacks on end devices haven't even been upgraded well enough to handle speeds above 1G bit/sec. McDonnell predicts that we will see experimental versions of 10G Ethernet gear within the year, but says wider deployment is two years or more into the future.

While the need for high speeds is a focus for larger networks, it may not be critical right now in smaller networks. According to Ingraham, many customers currently are using lower-end servers for successful iSCSI connections over Fast Ethernet links. Servers that are not supporting intensive storage access have no need for faster speeds. But as customers move to higher-performance servers, such as the next-generation InfiniBand-based servers that will deliver I/O bus throughput speeds over 2.5G bit/sec, the need for speed really matters.

But researching and developing ASICs that can switch traffic at 2G bit/sec speeds is "rocket science," and it's costly, according to Brocade's Totah. The challenge here is to attain a very high line speed without introducing latency and offer advanced capabilities, such as trunking and frame management.

Storage always has been expensive relative to Ethernet technologies, but pricing of Fibre Channel-based products has leveled off and will continue to decrease. According to Staimer, we can expect an average price of $750 per 2G bit/sec port on edge switches by year-end. This is down from an average price of $1,500 per 1G bit/sec ports in early 2001.

Staimer estimates that the per-port prices on director-class switches, which we define as those supporting 64 ports and higher, will be around $1,750 per port by the end of 2002. That's compared with an average of $3,100 per port on current products. Inrange's Lafferty projects that Fibre Channel pricing will continue to decrease at the rate of 10% per year over the next several years based on more widespread adoption of the technology.

Totah notes that before the events of Sept. 11, customers were buying a lot of stand-alone storage but not much of the connecting technology that would facilitate a SAN environment. Since then, the trend has shifted toward buying more SANs and a little less storage - with the goal of gaining more connectivity to make current storage more cost-effective. But Totah noted, too, that users should realize now that the return on SAN investment is high. If disaster strikes, the money you laid out for SAN technology will be well spent if that gear helps you get your business back online more quickly.

SAN buyer's guide index