Do SAS-based SANs make sense?

By Kevin Komiega

—There is a growing debate over a new way to build SANs using Serial Attached SCSI (SAS) as a fabric interconnect. The technology exists and does have potential benefits, but there is some question as to whether end users really need yet another type of storage-network fabric.

Building storage fabrics based on SAS is still just an idea on the drawing board. End users will get their first taste of SAS in servers with direct-attached SAS disk drives, as originally intended. But SAS is both a disk interface technology and an interconnect.

"I see the first round of SAS deployments being either parallel SCSI replacements or super-sized versions of Serial ATA," says Harry Mason, president of the SCSI Trade Association (STA) and director of industry marketing at LSI Logic. "But SAS technology does
provide a convenient interconnect that allows you to build interesting configurations."

Although SAS was originally designed to replace the aging parallel SCSI interface as an inside-the-box connection in direct-attached storage (DAS) implementations, it has many of the attributes of a network fabric. SAS can scale, allows multiple hosts to connect to multiple storage arrays, and offers the redundancy of multiple paths.

Where there is new technology there are always new ways to use it. Mason believes users and vendors alike will tinker with SAS to find out what problems it can solve. One immediate problem that SAS could alleviate is in blade-server cluster environments. "Parallel SCSI has been used in clusters for a long time, but only in fairly small clusters due to electrical and cabling limitations," says Mason. "Now, with blades bringing components closer together we can put together storage configurations rather neatly with SAS."

But with Fibre Channel and iSCSI already entrenched as the topologies of choice for SANs, throwing another fabric interconnect into the mix could muddy the waters and is dismissed by some as unnecessary.

To be clear, the pro-SAS movement is not gung ho in pushing SAS as a SAN interconnect. They merely believe the concept should be explored. Mason and other STA members have been grappling with a term that would define SAS-based fabrics without confusing users who have already accepted SANs as strictly Fibre Channel or iSCSI networks.

"There are some people who get heartburn over the idea of using SAS as a true fabric," says Mason. "We're clearly not trying to compete with long-distance technologies like iSCSI and Fibre Channel."

Avoiding end-user confusion is one of the main goals of STA. "Technically, a SAS fabric could be a SAN, but in a user's mind the term 'SAN' means a lot of things," says Linus Wong, a member of the STA board and director of strategic marketing for Adaptec. "In essence, you're implementing a SAN if you implement a shared-storage technology within a blade system."

The difference, Wong explains, lies in what is possible with SAS from a user point of view. "It's better to think of it as a cluster or local SAN," he says.

iSCSI has its place because it is low cost, Ethernet is already understood by the masses, and it works well in distributed environments, while Fibre Channel fabrics typically reside on the opposite end of the spectrum as relatively expensive, long-haul interconnects with high availability and performance.

Steve Rogers, director of product and technical marketing at Adaptec, believes that SAS may fit as a local, in-the-room SAN topology that is less expensive than Fibre Channel. He claims that a SAS fabric has much more bandwidth than an iSCSI SAN (although head-to-head performance comparisons would depend on which version of Ethernet is used for the iSCSI SAN, as well as other factors). Also, SAS is dual-ported, which allows users to take advantage of most of the high-availability and redundancy capabilities that Fibre Channel has, although at relatively limited distances.

Total cost of ownership may be the determining factor in driving adoption of SAS technology for SAN fabrics, although it's too early to determine the initial and ongoing costs required for SAS SANs.

SAS-based SAN fabrics can't be cobbled together yet due to a lack of adequate switching capabilities. SAS requires the use of expanders to allow for a host to connect to more than one device. The arrival of low-cost, high-port-count "fan-out expanders" is still at least
nine months away. The expanders will allow connection of multiple arrays to a SAS fabric.

Rancho Technology is one company that is producing 12-port, 3Gbps SAS fan-out expanders today for use in connecting targets and initiators. Rancho plans to produce a new crop of SAS expanders with up to 36 ports. According to Rancho's product description, the 36-port expanders are designed for high availability and scalable server clustering environments and front-end storage subsystems used in clustered, SAN, and NAS environments.

All of the components will soon be readily available for building SAS fabrics, but the need for them is still unclear to some.

"The only reason to bring in new technology is because it's either cheaper or better. And I don't think we really need another SAN fabric because Fibre Channel keeps getting cheaper and iSCSI is the ultimate in cheap," argues Steve Duplessie, founder and senior analyst at
the Enterprise Strategy Group research and consulting firm. Duplessie believes that today's cost and performance requirements can easily be met by either iSCSI or Fibre Channel fabrics.

Nevertheless, building a SAS fabric to implement what can loosely be defined as a SAN may soon become a viable option for networked storage, although the practicality of dedicating time and resources to the concept remains to be seen.

But some in the vendor and end-user communities are always looking for a new Mt. Everest. It could be that they will build SAS fabrics simply because they can.

This article was originally published on September 19, 2005