Sun Microsystems, DEC and Hewlett-Packard Co were among the scores of companies that cut their Unix teeth on the University of California at Berkeley variant of the AT&T Bell Laboratories operating system. The contribution made by Berkeley to the take-up of Unix, particularly in the scientific and engineering world cannot be under estimated, and there is life in the old dog yet, as Berkeley pre pares the up-coming BSD 4.4 for release: William Fellowes reports.

The next version of Berkeley Unix, BSD 4.4, which is due for release at the beginning of next year, will include Open Systems Interconnection protocols as part of an OSI-Posix project instigated by the US National Institute of Standards and Technology, (previously the National Bureau of Standards), and backed by DARPA, the US Defense Advanced Research Projects Agency. Both Transmission Control Protocol/Internet Protocol, TCP/IP, and Open Systems Interconnection protocols are being integrated into new BSD Unix features being put together at the University of Berkeley in California by a team led by Keith Sklower. These will add certain structures and protocols to TCP/IP to make it look like X25, and is being built on top of the Open Systems protocol development work by Marshall Roades of the Warnock Group.

UK involvement

University College London is contributing work on the X400 message handling protocol, the University of Nottingham’s work on the X500 address – per – user – worldwide X400 directory is to be included, and the University of Michigan will add other work on the TCP/IP and Open Systems protocol sets. Whether these Posix compliant additions will be included in future versions of AT&T Unix Software’s Unix as well as BSD, which are both Posix stamped, is unclear, because communications protocols in System V are treated in a slightly different way from those in the BSD variant. However the OSI-Posix project material will be put close to the public domain, and is hoped to foster the commercial application of Open Systems protocols. Sklower thinks that Hewlett-Packard Co may be interested in picking up these developments for its HP/UX operating system, although DEC already has its own DECnet system, and Sun Microsystems Inc has already implemented some of the ideas, (see below). What is clear is that the project should provide a springboard for the Open Systems protocol suite to be more widely adopted. It must be remembered that the inclusion of TCP/IP protocols in BSD Unix helped to make TCP/IP a success, and the inclusion of both TCP/IP and OSI under the same operating system should make the transition from TCP/IP to OSI easier – even IBM has adopted TCP/IP in its general Open Systems Interconnection product offerings as a stop-gap to full OSI. The Berkeley project itself is part of a more general effort going on at the moment to push forward development and integration of communications protocols. In particular Dan Jacobsen, just down the road from Berkeley, has devised a number of additions to the TCP/IP suite, generally known as the Jacobsen extensions, but probably more accurately described as a ‘tweaking’ of the control and transport layers. They enable 98% of the theoretical bandwidth on Ethernet to be used, as well as enabling communication congestion to be noted so that appropriate backing-off action can be initiated if needed. By installing one element of cache memory on each network layer, the system can also predict what is going to happen next on the stack, and again take appropriate control action. The first commercial implementation of Jacobsen extensions has been done by Sun Microsystems on its Sun-4 workstation. They enable data to be transferred off a Sun-4 at the rate of 8M-bits-per-second, dramatically improving upon the standard TCP/IP rate of 1M- or 1.5M-bits-per-second, and nearly as good as Sun’s own Network File System at 10Mbits-per-second. The performance was achieved by removing duplication and eliminating redundancy in the Internet Protocol element, and by reformulating the design of Lance chips in Ethernet boards. The p

opular TCP/IP protocol suite was developed in the 1970s by the US Defense Advanced Research Projects Agency as part of the effort to create the pioneering Arpanet, which links the big computers at all the major US research institutes and universities and proved most of the techniques that are widely used in today’s packet networks including such things a electronic mail, although Arpanet people called it Network Mail.

Error checking

The US government began to require TCP/IP in its networking contracts and the development of more powerful personal computers that could handle the processing required by complex protocols gave TCP/IP a boost. TCP is the transport layer protocol and IP is the network layer protocol, operating at layers four and three respectively of the seven-layer Open Systems Interconnection protocol stack. TCP/IP was originally developed for wide area networks, which tend to be less reliable than local networks, so its transport layer does more error checking, slowing down the transmission rate, but it has nevertheless become a de facto standard in networking technology. According to a study of network architectures by local network specialists Interlan Inc, Boxborough, Massachusetts, recently acquired by Racal Electronics Plc, TCP/IP’s major competition will come from this OSI de jure standard for interoperability, but the Berkeley project seems to suggest that some sort of co-existence is likely, especially in the near future. The International Standards Organisation has been gradually developing protocols for the seven layers of the its Open Systems Interconnection model, and now virtually every computer company has expressed support for the OSI standards. And the US government has now adopted OSI in place of TCP/IP through its implementation of GOSIP, the Government OSI Profile.