A multicomputer approach to non-numeric computation (multi-processors, database machines, parallel, algorithms, computer architecture)
Chaitanya K. Baru · 1985
In this research we study the architecture of a dynamically partitionable multicomputer system with switchable main memory modules (SM3) for non-numeric computations. The architecture supports efficient execution of parallel algorithms for database processing by (i) allowing the sharing of switchable main memory modules between computers, (ii) supporting network partitioning, and (iii) employing global control lines to efficiently support inter-processor communication. The data transfer time is reduced to memory switching time by allowing some main memory modules to be switched between processors. Network partitioning gives a common bus network system the capability of an MIMD machine while performing global operations. The global control lines establish a quick and efficient high-level protocol in the system. The network is supervised by a control computer which oversees network partitioning and other global functions. A simulation study of some commonly used database operations, using discrete-event simulation techniques, has been carried out and the results of the study are presented. Certain aspects of the system architecture have been modified on the basis of simulation results. The general-purpose nature of the SM3 system allows the implementation of a variety of parallel algorithms for database processing. We present a methodology for representing and transforming such algorithms in order to obtain the most efficient mapping of algorithms onto the underlying architecture. The methodology suggests a stepwise decomposition of the algorithm in order to exploit and parallelism. The extent of vertical parallelism is determined by the structure of the algorithm, whereas the extent of horizontal parallelism is determined by the distribution of data. In the transformation process, special attention is paid to the data movement and transformation aspects of the algorithm.