After all the coverage given to Serial ATA (SATA), it is now the turn of its younger sibling. Yes, Serial Attached SCSI (SAS) is under development for release next year, although some in the storage business are already questioning its viability and chances of success.
SAS is the "next logical step for SCSI," according to Linus Wong, a strategic marketing director at Adaptec, which recently tested a prototype SAS chip peaking at 5Gbit/sec. Wong says that the evolution of parallel SCSI will end at Ultra-320 - as far as he knows, no-one is developing targets or initiators for the proposed Ultra-640 generation.
Like SATA, Serial Attached SCSI is a evolution of the original, developed to run over a serial cable. It is the responsibility of ANSI's T10 committee, which is also responsible for the parallel SCSI standards.
The first generation of SAS specifies a 3Gbit/sec connection, with successive versions doubling this. Although 3Gbit/sec is roughly the same as Ultra-320's 320MB/sec, SAS is a point-to-point connection instead of a shared bus, so each device gets 300MB/sec of dedicated bandwidth, making it more scalable.
A further point in its favour is that SAS and SATA share the same cabling, and a SAS controller will include SATA by default - SAS is in effect a superset of SATA. SAS developers such as Adaptec argue this will bring great flexibility for users and system builders.
The SAS protocol goes beyond SATA in several areas, among them being support for wide links. This means you can aggregate SAS connections between the host and subsystem. For example, four channels could be combined to give a total of over 1GB/sec.
In addition, SAS defines dual-ported drives and can have both ports active. Another feature of the spec is the ability to use expanders to cascade multiple devices onto a single connection or a wide link, creating a sort of storage network with a theoretical limit of 16,256 devices.
SATA has similar functionality, but where SATA port-multipliers operate more like Ethernet hubs, with shared bandwidth, SAS expanders resemble Ethernet switches, with each device getting its own dedicated 3Gbit/sec of connectivity.
However, not everyone sees a rosy future for SAS. According to Ian Keene, Western Digital's European SATA business manager, the risk is timing and the resulting need for SAS to find a niche between better established technologies.
"SAS will be later to market than SATA, so the low end has already gone," he says. "There's Fibre at the high end, SCSI still in the middle, and we can do the same mechanisms with SATA. So if you can do some SAS features on SATA, such as dual-channel, do we need SAS?"
Although WD got out of the enterprise SCSI market a few years ago, it has had considerable success recently with high-end SATA products such as its Raptor hard disk range. Keene says that its lower cost SATA drives have been doing well too.
"We can do today what SAS does with our two SATA product ranges," he adds. "What's SAS going to bring?"
In his view, much will come down to cost. "Raptor is about twice the cost of parallel ATA, or 30% less than SCSI," he says. "Bulk SATA costs around 20% more than parallel ATA. SATA still carries a $10 [per drive] premium, that's too much for the PC market but OK for the enterprise."
He notes too that on the performance front, the next generation 300MB/sec SATA-2 offers more bandwidth than most applications can use. WD and others will still bring SATA-2 to market though, because it offers other advantages than bandwidth - some of them duplicating features of SAS, such as command queuing and dual-channel capability.
SAS supporters such as Linus Wong are unfazed, however. He says the ability of a SAS controller to also support SATA drives gives companies such as HP considerable freedom in how they build storage subsystems.
"Transactional data is still growing but with a huge growth in reference data, for example email, cheque images or medical images," he says. "There's a whole bunch of new applications storing those - it's a different type of data, it's larger files and they're accessed less frequently, so you optimise for capacity, not I/O speed.
"The challenge is how do you address both needs. Today, the system builder has to build two systems, but tomorrow they will be able to build one backplane and populate it with SAS or SATA, depending on the workload. SATA is getting more functionality, but fundamentally it is still a desktop drive."
The next step for Adaptec is to turn its prototype into a finished chip and test it with prototype drives from the likes of Seagate and Maxtor. Following that, the SCSI Trade Association plans a March plug-fest at the University of New Hampshire. Wong adds that the target is to ship product to OEMs in the second quarter for qualification, with volume production by the middle of 2004.