Work

Scheduling for memory management and prefetch in embedded multi-core architectures

Year: 2011
Type: dissertation
Abstract: This PhD thesis is devoted to the study of several combinatorial optimization problems which arise in the field of parallel embedded computing. Optimal memory management and related scheduling problems for dataflow applications executed on massively multi-core processors are studied. Two memory access optimization techniques are considered: data reuse and prefetch. The memory access management is instantiated into three combinatorial optimization problems. In the first problem, a prefetching strategy for dataflow applications is investigated so as to minimize the application execution time. This problem is modeled as a hybrid flow shop under precedence constraints, an \mathcal{NP}\text{-hard} problem. An heuristic resolution algorithm together with two lower bounds are proposed so as to conservatively, though fairly tightly, estimate the distance to the optimality. The second problem is concerned by optimal prefetch management strategies for branching structures (data-controlled tasks). Several objective functions, as well as prefetching techniques, are examined. In all these cases polynomial resolution algorithms are proposed. The third studied problem consists in ordering a set of tasks so as to minimize the number of times the memory data are fetched. In this way the data reuse for a set of tasks is optimized. This problem being \mathcal{NP}\text{-hard} , a result we have established, we have proposed two heuristic algorithms. The optimality gap of the heuristic solutions is estimated using exact solutions. The latter ones are obtained using a branch and bound method we have proposed.
Language: English
Open Access status: green
OpenAlex ID: W157059351
Cites: 49
Cited by: 1
Related to: 10