Parametric Micro-level Performance Models for Parallel Computing

Parametric micro-level (PM) performance models are introduced to address the important issue of how to realistically model parallel performance. These models can be used to predict execution times, identify performance bottlenecks, and compare machines. The accurate prediction and analysis of execution times is achieved by incorporating precise details of interprocessor communication, memory operations, auxiliary instructions, and effects of communication and computation schedules. Parameters are used for flexibility to study various algorithmic and architectural issues. The development and verification process, parameters and the scope of applicability of these models are discussed. A coherent view of performance is obtained from the execution profiles generated by PM models. The models are targeted at a large class numerical algorithms commonly implemented on both SIMD and MIMD machines. Specific models are presented for matrix multiplication, LU decomposition, and FFT on a 2-D processor array with distributed memory. A case study is done on MasPar MP-1 and MP-2 machines to validate PM models and demonstrate their utility