According to Unisys Corp sources, Novell Inc has signed a contract with Unisys to take the Single System Image facility, used as part of its distributed Unix operating system running on its recently launched Opus scalable parallel processing system. The Single System Image, somewhat similar in functionality to Locus Computing Corp’s Transparent Computing Facility – one of those interesting technologies that never really caught on (CI No 2,497) – is the most significant feature that differentiates the Unisys offering from IBM Corp’s SP2 massively parallel processing system (CI No 2,651). Although IBM mentions a single system image capability in its marketing brochures, it has not yet implemented one, says Unisys. Single System Image enables users to log in one time and to allocate access to individual areas at one time. On an IBM SP2, claims Unisys, such tasks as allocating a password to a new user can mean doing the same thing individually on every single node. Single System Image works by using virtual memory to store information about all the activities on each of the nodes, and keeping the inode and fnode look-up tables updated. Performance overhead is not a problem, says the company, due to the high speed 175Mbps mesh technology it bought in from Intel Corp. There are two ways of partitioning an Opus system: software partitioning, which enables the option of a breakthrough; and hardware partitioning, where you’re not allowed to break the barrier. In this case, Single System Image will present just that, on one partition, and communicate with other partitions by standard means, such as Network File System. Along with Single System Image, Unisys also implements automatic load-balancing at the system level, which works in a similar way to symmetrical multiprocessing systems, enabling large tasks to be switched between processors automatically. It has also modularised Unix by breaking down the processes within Unix itself – so you can have a process manager, object manager, streams manager and IPC manager distributed across various nodes. Unisys began its development of distributed Unix back in 1989 in San Jose, and has been working closely with Chorus Systemes SA during that time on the Chorus/Mix System V.4 implementation. Three years ago it began joint development work with Novell and Chorus on the Amadeus project, designed to look into the modularisation of Unix and its integration with microkernels. Unisys says it also helped with the development of the multiprocessing extensions to System V.4. None of this means that Novell will end up taking the Unisys code as part of its projected SuperNOS project, despite all the similarities – although it well might. All that Novell would say is that some of the Unisys work might not be appropriate to achieve its primary goal – the full integration of UnixWare and NetWare technologies.
