2.3 The Parallel SCSI Bus

The most common method for connecting disk and tape devices to file and application servers has been via traditional SCSI bus cabling. Although SCSI is really a block protocol, the term is also used to refer to the parallel cabling scheme that evolved with the protocol.

The original SCSI physical layer transport was a parallel cable with 8 data lines and a number of control lines. Transmitting 8 bits of data during each transmit clock provides a relatively high bandwidth, but electrical issues restrict the total distance allowed by most SCSI implementations to 15 25 meters. The SCSI parallel bus architecture has evolved over time, with higher bandwidth provided by wider data paths (16 data lines and 32 data lines) and faster clock rates.

One of the difficulties presented by parallel bus architecture is a phenomenon known as skew. If 8 or 16 bits of data are sent simultaneously in parallel, small differences in propagation delay along each data line may occur, and not all bits may arrive at the destination at precisely the same moment. Skew refers to the difference in arrival time for each bit comprising a data word. The skew window is the time difference within which all bits must arrive. As shown in Figure 2-7, the greater the differences in propagation delay, the wider the window must be to ensure that all data bits are captured. One factor in skew is the differential cable propagation delay between each data bit that is, the difference in the amount of time taken by a signal to travel from one end of the cable to the other, as a function of bit position. For a given cable length, an increase in clock rate requires a reduction in the skew window and the minimizing of differential propagation delay.

Figure 2-7. Timing skew on parallel bit transmission

Additionally, parallel SCSI requires termination of any unused ports. As multiple SCSI devices are daisy-chained together, end devices must be terminated to avoid erratic signal interference. Proper cabling and termination are critical for stable parallel SCSI operation. Marginal components or improper termination can cause data corruption or transaction failures.

File servers may house disk drives internally and provide either parallel SCSI cabling or a parallel SCSI backplane for installing disks. Current data requirements for most enterprises, however, make internal storage impractical.

To accommodate hundreds of gigabytes or terabytes of data, it is necessary to deploy multiple external disk enclosures, which can be connected by parallel SCSI cabling to multiple SCSI host bus adapters installed in the server, as shown in Figure 2-8.

Figure 2-8. Server with multiple SCSI adapter cards installed

The movement of data from internal storage to external storage has been a driving force in finding alternative solutions to parallel SCSI cabling. For enterprise-class storage requirements, the restrictions in terms of distance and number of devices supported on a SCSI bus make it difficult to support larger storage capacities. The primary weakness of parallel SCSI for enterprise applications, however, is that it binds storage resources to a single server. Storage resources cannot be shared easily by multiple servers, and unused storage capacity behind one server cannot be utilized by another. For enterprises that must support and manage tens or hundreds of terabytes of data, a networked storage solution is needed.

For direct-attached storage configurations, the limitations of parallel SCSI are being addressed by Serial Attached SCSI (SAS). The SAS specification is being developed by ANSI T10 as the next-generation direct-attached storage interface. Although SAS is not intended as a replacement for networked storage, it will provide efficiencies in local storage connectivity and will be immune to the skew and distance limitations of parallel cabling.