With Intel Corp boss Andy Grove offering support for what he believes is a high-end Unixification process, Unisys Corp has issued a clarion call for the industry to rally round its effort to create a standard commercial Unix for large-scale systems. Unisys has rolled out Unix System V.4.0 components implemented on top of the Chorus Systemes SA microkernel as Opus for its Open Parallel Unisys Servers, which are based on Intel’s Scalable Parallel Processors (CI No 2,652). As well as offering the initial Opus release OEM, Unisys is also implementing the work on its Pentium-based U6000 Series symmetric multiprocessing servers, which at the high-end are re-badged Sequent Computer Systems Inc Symmetry 5000s, and its own architecture lower down. There are no takers yet, except Intel, which will resell the Opus package, and Unisys has not decided whether to retain the Opus name for the symmetric processing units.

Componetised

AT&T Corp, Pyramid Technology Corp, Sequent and Tandem Computers Inc have each, at one time or another, called for a standard approach to high-end commercial Unix, to redress Novell Inc’s lack of appetite for meeting large-scale Unix requirements. The gist of what we heard goes something like this: right now Opus is a componentised Unix System V.4.0 on top of a Chorus microkernal, splitting functions such as process management, file management and streams – which can be distributed across nodes – under a single system image. Add to that Novell’s Unix System V.4.2 MP and Chorus’s telcommunications work – development already under way with at least two other partners that have chosen to remain anonymous (AT&T and Sequent swore to us it wasn’t them). Throw NetWare into the pot, again cloaked in a single image, and run the result across mixed parallel, symmetric processing and workstation environments. That brings us to ground where, at least in theory, SuperNOS should be growing. Meanwhile Chorus and Unisys, who have been working together on Unix System V.4.0 since at least 1991, are also both involved in the Esprit III-funded Ouverture project, which is apparently feeding core microkernel technology into SuperNOS. Chorus expects to be talking about the results of the Ouverture work over the next few weeks. Although Unisys claims to have originated its unique Single System Image software featured on the Opus boxes, we couldn’t help noticing that Intel’s Scalable Systems Division included something of exactly the same name and providing pretty much the same functionality when it launched its latest 80860-based Paragon, the XP/S MP, in October. Intel’s version enables its machine to appear as a single system with a single process ID space and file system and tra nsforms a network of independent Unix processors with no shared memory into a unified Unix environment. Application programmers and system programmers are freed from the details of locating critical resources such as remote files, system processes and input-output ports. The Unisys version, on the other hand, makes all processors on the system visible to each other, and has the ability to view and manage all system resources – such as memory, storage, printers, users and communications devices – as though they were part of a single system. Unisys maintains that the two are completely unrelated, and a spokesman said that Unisys had separately invented its version – that is it wrote the code.

By William Fellows

Admittedly, the Unisys system also includes automated load-balancing, though applications must be tweaked for the load-balancing to be fully optimised. Of course Unisys Corp already has a decision support box – the DataCentral enterprise database server, launched back in March 1994. Built around a proprietary 2200 Series mainframe, the strength of DataCentral is that it combines decision support with high volume transaction processing, something the new Opus box is rather weak on. While DataCentral appears to have little future beyond Unisys’s dwindling 2200 Series mainframe user base, some of th

e expertise is very likely to filter down to the Opus development in order to improve transaction processing performance there. Both Digital Equipment Corp and Sequent favour the clustering together of symmetric processing systems over scalable parallel systems or massively parallel processing machines. DEC says it can cluster up to 96 8400 AlphaServers together, each with up to 12 Alpha chips rated at 550 SPECfp inside, giving more flexible system offerings at a fraction of the price. They offer a common data view, and with the recently announced Very Large Memory facility, can access up to 12Gb of an Oracle database directly from main memory. Decision support is not about number of processors, it says, it’s about how fast you can get information in and out, and memory is much faster than disk. Sequent says that while a massively parallel processing system might be balanced as an entry-level system, adding new nodes in the future means that you are forced to upgrade input-output, the processor and memory together, which can make it a very expensive option if all you need is more input-output. As we reported, Sequent’s collaboration with Intel will result in a Video Pum p, a specialist scalable parallel processing Media Server for the video-on-demand market that will always be sold in conjunction with a Sequent symmetric processing applications server acting as a front-end, along with Asynchronous Transfer Mode switching technology. Sequent is working direct with Intel’s Scalable Systems Division, its neighbour up in Beaverton, Oregon: it is not taking any of the Unisys-developed Opus stuff, but is using the same underlying two-dimensional mesh technology for its P6-based machine. Unnoticed (at least by us), Intel’s Scalable Systems Division discontinued work on its Hypercube-based parallel supercomputers using the iAPX-86 line of processors some time ago, saying that two-dimensional technology used in the 80860-based Paragons turned out to be the more efficient approach.

Opus for dough

When asked how soon an iAPX-86 Paragon line will emerge, the company now says that it’s a reasonable assumption that one is on the way. Unisys spoke rather optimistically – we think – of having Windows NT up on Opus within a year. It claims Intel has done a whole bunch of work and made good progress parallelising Windows NT. Unisys folks say their company needs a firmer business relationship with Microsoft before NT could appear on Opus. Ed Masi, vice-president and general manager of Intel’s Scalable Systems Division, formerly the Supercomputer Systems Division, is resigned to replacing the OSF/1 kernel that has been running on its Paragon technical scalable parallel processing engine as OSF/1 slips away. Intel has already re-written the problematic Mach-based inter-process communications technology found in OSF/1 for its own purposes. It says it does not believe the Unisys Unix System V.4/MK communications and node-linking stuff is robust enough for its requirements, although it quips OSF/1 for show, Opus for dough.