The Network Computing System announced by Apollo Computer Inc last week (CI No 620), could turn out to be one of the most significant networking announcements in recent years, and be the trigger that turns local area networking into something much more than it for the most part represents today – a re-implementation of the concept of the multi-user system of the 1970s with a little more local intelligence added at the workstations. And it is exactly what IBM should have come up with three or four years ago to clean up the morass in the more scientific and technically-oriented side of its mid range. Whether the Network Computing System – which has been put into the public domain by the Chelmsford, Massachusetts company – will do Apollo any more good than Ethernet has done Xerox, Unix has done AT&T, is another matter. If it really works as Apollo claims, the company will deserve all the success that comes to it – but the market is notoriously cruel to genuine innovators. Apollo describes the Network Computing System as the first commercially available set of distributed computing products for developing and running application programs across networks of incompatible computers from multiple vendors. Apollo believes that it is the first system that can distribute modules, or parts, of a single application program to the specialised computers on the network that are best suited to execute it efficiently – artificial intelligence engines, database machines, automatic test equipment, simulators, parallel processors, and supercomputers. It is also conceived to find idle computers and keep them busy when other parts of the network are becoming overloaded – just as a good mainframe operating system does when running a mix of batch and interactive applications – but even a main-frame with integrated vector processor only scratches the surface of the capabilities of specialist machines such as a Symbolics 3600 artificial intelligence computer, a Teradata parallel database machine, a Convex C-1 mini-supercomputer, a Floating Point Systems T-100 Transputer supercomputer, a handful of Apollo node processors and a string of AT-alike terminals. The Network Computing System is designed to lay a firm foundation for the development of dynamic, integrated technical applications whose tasks can be run productively on remote systems right from the workstation, without compromising interactivity. The company gives as an example a Network Computing System-based mechanical and electronic design applications can enable an engineer to capture and display graphically a design on the workstation while a background supercomputer concurrently analyses the design. The workstation could then display the results graphically in real time. In addition to unveiling generic tools for developing applications between dissimilar computers, Apollo introduced Network Computing System source code to run under Unix and DEC’s VAX/VMS. Network Computing System, needless to say written in C for maximum portability – C is also highly regarded as a system software programming language – is described as open and portable system. The source code can be licensed, is fully documented, and Apollo has published the specifications so that others can implement the system without having to pay Apollo for the privilege. The system is based on industry-standard networking protocols, including TCP/IP for Ethernet and DECnet, Apollo’s DDS Domain Distributed Services, IBM’s SNA, and MAP/TOP, and Apollo sees it complementing distributed file systems such as Sun’s Network File System, AT&T’s Remote File Sharing, and its own Domain, by providing the computational sharing that is absent from those systems.
The system consists of three components that the company reckons solve the problems that previously prevented the development of true inter-vendor network computing – a Remote Procedure Call Run-time Environment – transparent to application programs – that handles packaging, transmission, reception of data, and error correction between the client and the parts of th
e application on the users’s workstation and on the computers providing remote services; Network Interface Definition Compiler, which compiles Apollo’s new high-level Network Interface Definition Language, NDL (shouldn’t that be NIDL, especially as Unisys-Burroughs already has an NDL?), into portable C source code that runs on both sides of the connection; and a Location Broker, which enables applications determine during program execution which remote computers on the network can provide the required services to the user’s computer. The NDL Compiler costs $1,000 per workstation or $8,500 per site; the NDL Source Code, supplied on tape, is $25,000; the Network Computing System Unix Run-time Source Code and the VAX/VMS Source Code are $1,000 each; for those who want to do their own thing, the Network Computing Architecture Public Specification is just $80; and the Apollo Specific Network Computing System Documentation and Runtime Source Code is $250. All are set for third quarter delivery.