Energy-Efficient Computing over Streams with Massively Parallel Architectures

Sammanfattning: The rise of many-core processor architectures in the high-performance computing market answers to a constantly growing need of processing power to solve more and more challenging problems such as the ones in computing for big data. Fast computation is more and more limited by the very high power required and the management of the considerable heat produced. Many programming models compete to take prot of many-core architectures to improve both execution speed and energy consumption, each with their advantages and drawbacks. The work described in this thesis is based on the dataflow computing approach and investigates the benets of a carefully designed pipelined execution of streaming applications, focusing on particular on off- and on-chip memory accesses. We implement classic and on-chip pipelined versions of mergesort for the SCC. We see how the benets of the on-chip pipelining technique are bounded by the underlying architecture, and we explore the problem of ne tuning streaming applications for manycore architectures to optimize for energy given a throughput budget. We propose a novel methodology to compute schedules optimized for energy eciency for a fixed throughput target. We introduce Schedeval, a tool to test schedules of communicating streaming tasks under throughput constraints for the SCC. We show  that streaming applications based on Schedeval compete with specialized implementations and we use Schedeval to demonstrate performance dierences between schedules that are otherwise considered as equivalent by a simple model.