Hey, you. Yes, you, with the fat wallet and the taste for state-of-the-art desktop storage technology. Hard disks are becoming more sophisticated, but were you aware that the technical specifications around them are becoming ever more fragmented? Deep in the heart of the US, an ANSI committee containing representatives from a host of companies is working on a new SCSI standard – SCSI-3 – which is designed to blow the socks off the SCSI standards that have gone before. The problem is that this time, the committee has altered the process used to define that standard, making it much more open to interpretation by different vendors. This means that when, as a user or an OEM customer, you buy a piece of hardware advertising compliance with the SCSI-3 standard, you may have to conduct research to find out exactly how it’s compliant and with which parts. The SCSI-2 standard is a one-document standard. It’s still complex, with 80% of it labelled as optional, but at least there’s a core subset of SCSI-2 specifications that manufacturers can follow, to ensure compatibility. The ANSI committee responsible for SCSI-3 has separated the standard into documents that handle different levels of operation, in turn separating these documents into a number of optional documents that are more often than not mutually exclusive. This hierarchical approach to standards definition has the drawback of creating a mix-and-match effect, which means that the standard implemented by one manufacturer may be totally different from that used in another’s product – and yet they can both call themselves compliant.

Glue logic

The standard’s top level consists of three documents. The first handles the physical protocols, defining what the data gets sent along, and at what speed, and that sort of thing. The second document handles the logical side of data transfer – exactly what gets sent where and in what order. It provides the glue logic that relates software commands to physical action. The third and final top level document handles the command set and as such is the easiest to put together, because it deals with software rather than hardware. Going down a step reveals the documents’ component parts. Within the first, physical, document there are five sub-documents. First is the SCSI Parallel Interface document which was completed last summer. This deals with the parallel interface, taking SCSI-2 standards and defines how to operate the fast transfers associated with this standard over single-ended electrical cable. SCSI-2 handles only the more expensive and less common differential option. The Parallel Interface document for SCSI-3 uses a 68-pin connector for wide SCSI to increase the bus width from 8 bits to 16 bits, upping the transfer speed. The next three handle different types of serial interface for sending data over paired wires, while the last one handles the IDE cable, which is the conventional AT-based cable, used to send different commands. The second, low level logic document, also handles the three serial interface formats.

By Danny Bradbury

The first is the 1394 standard, which is also known by some as Firewire. This is an IEEE standard for use in desktop SCSI applications. The second is the Serial Storage Architecture, developed by IBM Corp in the UK as a proprietary interface for use in disk array applications, and finally Fibre Channel, which was first conceived about four years ago as a way of running the SCSI standard over fibre optic cables to supply minis, mainframes and supercomputers. People are starting to change from copper cable to Fibre Channel now, to increase the speed and distance available. The whole idea behind producing a standard based on these different standards is to make the standard as scalable as possible, so that it pertains to all levels of user, from desktop personal computer users to supercomputer owners. The documents within the command set are broken up into handling different devices – there is one for block devices, and one for streamer-based units such as tape drives. There is one for multimedia devic

es and one piece of the standard especially designed for graphics storage devices. Finally, the committee has defined a piece of the standard for controller devices, such as those used in disk arrays. The whole caboodle is pulled together by a document called the SCSI Architecture Model. The document is a sort of standard for standards, and governs how the various components of the three top-level documents interrelate and communicate. The Architecture Model document should have gone through approval in three to four months, along with the low level documents for logical processing. By the end of 1994, everything should be through apart from the command set documents, and these are the easiest things to define. What is interesting is that there is already talk of an SCSI-4 standard, which is being driven by the likes of Digital Equipment Corp and Sun Microsystems Inc: the two are driving standards forward by bringing sexier technology to the table. DEC is said to have suggested to the committee that it quadruples the 10M-byte and 20M-byte per second parallel transfer rates in the existing SCSI Parallel Interface document. This would probably form the basis for a second generation version of the document. A council representative told Computergram recently that work on the parallel interface specification for SCSI-4 would more than likely start before some of the SCSI-3 documents are finished. Meanwhile Sun is said to have proposed a cable-less connection standard for SCSI, which would enable users to slot disk subsystems directly into their machines.

Core subset

Wherever there was a difference between the SCSI-3 and SCSI-2 specifications – such as the doubled bus width in the SCSI-3 parallel specification – the council took pains to make the latest standard backwards-compatible with its predecessor. In the case of the bus width, a 16-bit SCSI-3 device plugging into an 8-bit SCSI-2 bus, there has to be a change made to the cabling to overcome the problem. Making the SCSI-3 standard easy to adopt is a little harder, though – pick and mix standards make the marketing to end users and even OEM customers potentially more difficult, and at least one representative on the council was painfully aware of that. The SCSI-2 standard is arguably similar, but at least there, OEM customers have a core subset that they could use to ensure compliance. SCSI-3 on the other hand consists of completely interchangeable components. Maybe it is to be expected that any enabling technology that is open to contributions from a number of third parties will start to fragment as it evolves. It creates a marketing nightmare, but from the technical point of view it can only be a good thing.