System architecture and design using co-operating groups of real and abstract components

Many techniques long used for the partitioning and packaging of hardware are gradually being adopted for software with the move to using cooperating self-contained software components. There has been a trend over the last decade towards implementations based on multiprocessor distributed concurrent systems with software components allocated to processors. Designing the system-wide operation, synchronisation and exception handling for such a complex system becomes impossible at the level of traditional process synchronisation primitives, as it becomes intellectually unmanageable to grasp the system-wide behaviour when tackled piecemeal and `bottom up'. An alternative top down approach to structuring and designing the overall behaviour of highly concurrent systems implemented using components is presented. It is based on using one or more simple abstract state machines to define the system-wide synchronisation required in terms of the desired system-level behaviour rather than in terms of the low-level synchronisation primitives chosen for the fine grain of the implementation solution. A practical example involving the creation of a high availability processing system made up of several commercial quality computers is used to illustrate the concept.