Multiprocessor Performance

Part 1 Multiprocessor architecture, program structure and machine performance: performance potential and limitations of parallel computer architectures; performance issues in some typical multiprocessor architectures; different types of multiprocessors. Part 2 Multiprocessor speed-up and Amdahl's law: an amendment to Amdahl's law; a model of intrinsic programme behaviour. Part 3 Performance of interconnection networks: bandwidth of the Baseling and Omega networks with the DROP approach; numerical solution of the DROP model; bandwidth of the DROP model; the HOLD network. Part 4 Parallel programme performance - series-parallel programme structures: the SPTG model of programme structure; distribution of programme execution or completion time; numerical solution for the execution time distribution function. Part 5 General acyclic random graphs as models of parallel programmes: asymptotic properties of the speed-up; asymptotic analysis of the mean processing time; comparison of the approximation with simulation results; maximum likelihood estimation of p the probability of precedence between tasks; task graphs with communication times between tasks. Part 6 Multiprocessor performance with task-graph models: the multiprocessor system model; derivation of the job response time; numerical examples and model validation; extension of the method to the analysis of multiprocessor systems with different types of parallel programmes; the case of systems with processors of different types. Part 7 Supercomputer performance evaluation: performance of a single processor; performance of supercomputers having multiple vector processors; supercomputers with a limited number of vector processing facilities; Los Alamos benchmark characteristics. Part 8 Performance analysis of the connection machine: ROUTER network performance; highly balanced computations; the connection machine with floating point processors.