Hewlett-Packard Co this week unveils its HP-UX 10.0 Unix operating system, unifying the different 9.05 and 9.04 strains that have been running on its Precision Architecture RISC Series 700 workstations and 800 servers respectively. It was originally planned for sometime last autumn, but Hewlett-Packard’s giving 10.0 the ‘push-pull’ routine, sending out a new business developer release this month – minus some key functionality – while the installed base gets to see the completed system software, if it asks for it, from mid-year. It is the way customers have asked for it, the company maintains. Although Hewlett-Packard is promising full backwards compatibility for 9.0x applications via new features such as Fast Transition Links, which intercept old directory path names and redirects them to the new System V.4 file system layout, there are only some 50-odd applications, including the major databases, that will run under 10.0 from day one, out of the 9,000 or so applications now claimed for 9.0x . Recompilation will be required to take advantage of 10.0-specific features, including the journalled file system, and optional OnLineJFS. But at least with Fast Transition Links, Hewlett believes it will avoid some of the compatibility problems that have dogged SunSoft Inc’s painful transition from SunOS to Solaris. True, up to a point perhaps, but then HP-UX 10.0 is only half out of the gate compared with Solaris 2.x. HP-UX 10.0 retains the Unix System V 3.2 kernel – albeit significantly re-worked with a System V.4 file system layout – and does not expose kernel-level threads for multi-threaded applications. Although it was never part of the game plan for this release in any case, Hewlett-Packard says it also looked at Sun’s experience with multi-threading and concluded that customers are simply not ready for it. There are few tangible performance benefits and nor are there the development tools nor applications to take advantage of the technology, it says, and Sun gets little pay-back because of the practical difficulties of using it. But Hewlett nevertheless promises just such kernel-level threading by year-end in a dot release of 10.0. That means yet another transition for independent software vendors and users if they want to take advantage of any performance benefits, it says. Sun reckons its implementation speeds system software, especially networking, by between 30% to 40%. By that time Hewlett will presumably have symmetric multiprocessing desktop hardware to take full advantage of it as well.
