The Open Software Foundation has started referring to OSF/1 as IK, short for integrated kernel, in contrast to MK, its new microkernel technology, which will be going into its first general release as MK 1.0 in the third quarter of this year. The pair will be developed along separate but related tracks until they converge in the summer of 1994 with the third microkernel iteration, known on the IK side as OSF/1.3. At that point, says the Software Foundation’s vice-president and resident technology wizard Ira Goldstein, his team will have solved the kit problem and produced a microkernel as suitable for the workstation as for the supercomputer. This hybrid or common release, in which reusable parts will carry over from IK to MK, will be a distributed real-time system, secure to the B3 level and configurable for uniprocessors, multiprocessors and supercomputers without current liabilities. (Right now, for instance, IK is faster than MK on a uniprocessor). Microkernel technology is particularly important for the next generation of massively parallel machines that have caught the industry’s fancy, Goldstein said, alluding to plans such as IBM Corp’s to produce highly parallel boxes out of the RS/6000’s Rios chips. About 12 companies including IBM, Groupe Bull SA and Siemens Nixdorf Informationssysteme AG, he indicated, have the MK snapshots and two major firms, apparently other than Intel Corp’s Scientific Supercomputer Division which is putting OSF/MK into its Paragon machine, but identified only as having revenues of over $100m a year, have committed to using the microkernel in future projects.
Ira Goldstein talks to Maureen O’Gara
As with today’s OSF/1, the basis of this MK 3 (or MK’94, as Goldstein decided to call it for clarity’s sake during our interview), will of course be Carnegie Mellon University’s Mach. The Open Software Foundation, however, is not bound to follow Carnegie Mellon mechanically, Goldstein said. For instance, it’s already looking at other technologies like X kernel, the University of Arizona’s framework for defining network protocols. Real time won’t spring full blown into the microkernel either. MK will begin to support real time working next summer with the release of MK 2 (MK ’93) which is meant to extend the microkernel from the computer to the cluster. Goldstein says that by embracing fibre optics over copper (which is too slow), a relatively ordinary workstation can be turned into a supercomputer. Goldstein’s team will pile as much functionality as it can into each of these prospective releases, with MK’93 running the most current libraries and commands. One of the issues MK was meant to resolve back at the height of the Unix Wars was the purge of AT&T code from future Open Software Foundation operating systems. Now that the white hot rage of those days is cooling the issue seems less lively. The microkernel could be purified of any Unix Labs code, Goldstein claims, but the question now is whether to bother. The code used is at the System V.2 level, and as Goldstein pointed out, any royalties accruing from it to Unix Labs are not dependent on the cost of the system. They would be trivial on a massively parallel machine. Whether they would be burdensome at the lower end remains to be seen. Goldstein suspects Unix Labs’s own microkernel strategy – if for no other reason than the arm’s length relationship it has fashioned with its microkernel choice, Chorus systems will not be able to keep pace with his. With the new administration now ensconced at Unix Labs and pragmatism its watchword, Goldstein wonders out loud whether it’s now time for them to put System V.4 on his microkernel.
