Abstract :
Distributed systems are incapable of handling certain problems which cannot be broken down into sub-problems without the use of communications bandwidths which such systems cannot provide. If this bandwidth can be provided, however, as in a scaleable massively parallel system then significant parallelism can be found in such applications. In particular Relational Database Management Systems (RDBMSs) have been shown to be capable of being parallelised in this way. Often it is then possible to encapsulate most of the difficult parts of parallelising a business application inside the RDBMS, leaving the remainder of the application as multitasking sequential code which can be distributed if necessary. Oracle was the first RDBMS vendor to recognise this and Parsys has been working with them since 1989 to implement their system, firstly on the T800 transputer and now the T9000. The author looks at how this has been acheived