Systematic Development of High-performance Distributed System Architectures

The author presents a view of how performance requirements are considered in the design of distributed and contributed systems, using the formal language LOTOS that does not support the specification of performance constraints directly in language elements. The design trajectory covered in this approach is discussed in terms of specific transformation steps that are relevant to the fulfillment of non-LOTOS requirements. The following four performance issues are addressed: (1) In the first formalization step, parallelism should be preserved as much as possible. (2) During the first transformation steps, process decomposition is performed in which knowledge about implementation elements is included. (3) The selection from among different structures can be supported by queuing networks for analysis and assessment for their relative applicability. (4) During the later transformation steps the implementation of abstract interfaces as defined in the specification should be efficient; the events should be implemented using appropriate call structures, and implementation mechanisms such as buffer management are relevant