SCSI: Past, present and future

The book of Genesis is filled with genealogies (not much of a surprise to any linguist), telling us that so-and-so begat so-and-so, and providing readers of the Bible with a patriarchal lineage that stretches back to Adam.  The genealogy of enterprise disk drives has its own lineage, although not nearly so distinguished or long-lived.

A case in point is the SCSI protocol, which has dominated open system enterprise shops for the last 17 years or so.  Most people have forgotten that “SCSI” is really an acronym for Small Computer Systems Interconnect.  This is probably just as well, as “small” really hasn’t applied to SCSI for more than a decade.  In fact, this protocol – originally designed for disks attached to stand-alone workstations – now provides data to the biggest machines in the largest enterprises.  SCSI has never been a very popular option for the PC desktop device but along with Fibre Channel-based devices, SCSI owns the high end.

The genealogy of SCSI is pretty much as follows: SCSI (5M bit/sec) begat Fast SCSI (10M bit/sec), Fast SCSI begat Ultra SCSI (nee Fast-20, at 20M bit/sec), Ultra SCSI begat Ultra 2 (40M bit/sec), Ultra 2 begat Ultra 80, which begat Ultra 160, which begat Ultra 320. As you can see, as the protocol got faster the nomenclature got more descriptive.

Of course, no version of the SCSI protocol had a lifespan approaching that of even the shortest-lived patriarch, but then SCSI was competing with the rival Fibre Channel technology and couldn’t afford to rest on its laurels. 

SCSI is about to spin out another, faster, version, but this time the begetting calls for a bit more explanation.

Until now SCSI and ATA disk technology (which owns the desktop) have had only one thing in common: both depended on parallel interfaces between the storage device (disk, tape, optical, etc.) and the controller that sat inside the computer or storage device.  Parallel worked well for almost 20 years, but now both SCSI and ATA are shifting to serial attachments. Why?

Serial SCSI (known as SAS – “serial attached SCSI”) enables a number of advances over its parallel predecessor.  The most obvious of these is a result of its much smaller cabling requirement.  The cable and connector are smaller, so significantly smaller form factor disk drives will be available, allowing vendors to significantly upgrade the storage density (in terms of number of devices) in each bay of a storage array. 

System builders are going to like it because the smaller cables are easier to route around the inside of a chassis, and the serial interface allows for much longer runs of cable than did the parallel interface. 

SAS will also allow up to 128 devices on a single bus.  This is significantly more than parallel, which allowed a theoretical maximum of 15 devices – 15 disks and a host adapter – but in reality never offered the bandwidth to support nearly-that many streaming devices on a single bus. Whether the ability to put all those devices on a single bus is a real advantage is problematic, unless throughput is not a consideration.

What is particularly interesting about SAS though, is the fact that SAS devices are going to be plug compatible with SATA devices, allowing SATA disks to be mixed in with SAS devices as the spirit (or budget) moves the admin.

Next time, we will look at who is supporting the new technology, when we can expect to see it, and perhaps even try to understand why we might want this stuff anyway.