* Reliability features of Serial Attached SCSI
Serial Attached SCSI drives are fast, robust, and for the most part identical to the high-performance devices already in use in the enterprise. Not only are they built to the same specifications that the current crop of parallel SCSI and Fibre Channel drives are, but they also make use of the same SCSI command set, have the same form factors and rotate spindle speeds.
In short, just about all drive electronics, mechanical components and media are the same; the only differences lie in the physical connector, a few added features and in the protocols at the PHY, link and transport layers.
So other than speed, what’s all the noise about? Much of it has to do with reliability.
The new generation of SCSI drives comes equipped with two ports per drive. Dual porting means added reliability, because now devices can be connected to a SCSI initiator (or to different initiators) through each port. Two connection points means that the major nemesis of high-availability systems – a single point of failure – has been eliminated.
Reliability is also enhanced due to better communication on the SCSI bus. The signals – both the data and the SCSI commands – are now sent serially rather than all at once, which yields other improvements that contribute to system reliability.
First, the likelihood that the various signals will interfere with one another (“crosstalk”), a major difficulty with any parallel data transmission, is essentially eliminated as the signals go one at a time.
Crosstalk was the major reason all those wide SCSI ribbon cables (or thick differential cables) we all have learned to loathe were so cumbersome. Every other wire in the cable was a grounding wire, inserted to protect the signals on the wires by isolating them from the wires on either side.
Reliability also improves because “signal skew” is eliminated. With any parallel data transmission, including parallel SCSI, signals must arrive at their destination within a sharply defined window (measured in nanoseconds), so timing is king. If a pulse arrives too soon or too late, the data get garbled. With parallel SCSI, this has meant that cables were cut to very precise measurements, and that the maximum distance between a SCSI initiator and a target was severely limited.
With serial transmissions, however, timing of the signal flow is much less complicated. Data and commands are sent one after the other, so the skew issue disappears. Also, because the line voltage of Serial Attached SCSI (SAS) is very low compared to that of parallel SCSI, the swings from high voltage to low can happen much faster. Since the signals themselves turn on and off more quickly, more time is available for data transmission.
Finally, because SAS devices use a point-to-point connection with their initiators (which is to say, one device per cable), disks don’t contend with one another for access to the host bus adaptor.
Simpler system design, simplified cabling, simple point-to-point connections, elimination of crosstalk and signal skew, elimination of any single point of failure through dual porting, and improved signal integrity all mean more reliable systems, higher data availability and better throughput.
So is the choice for you going to be SAS, Serial ATA, or Fibre Channel? That comes next time.
