Evaluation of Load Balancing Techniques
Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesis
Performing calculations that are complex require a large set of processors in order to achieve a faster solution time. When parallel calculations are performed there should be an increase of performance that corresponds to the number of processing units. The problem is how this can be performed efficiently on both shared and distributed memory and what consequences each method brings to the solution. Balancing the workload equally between the processes and the techniques to achieve this will therefore be of great value when complex calculations are preformed. A non-recursive implementation of the travelling salesman problem will supply the necessary complexity for the study and will also bring an interesting aspect to be solved in parallel. The method that is researched is based on both OpenMP for shared memory and MPI for distributed memory machines. The tests performed were done with the help of a hypercube implementation and a local work-pool for each process.The system was evaluated in a communicational aspect where the performance increased when communication decreased between processes. When running the travelling salesman problem in parallel results showed that partitioning of the data have a major impact in how fast a problem is computed in parallel.
Place, publisher, year, edition, pages
, UMNAD, 911
Engineering and Technology
IdentifiersURN: urn:nbn:se:umu:diva-58255OAI: oai:DiVA.org:umu-58255DiVA: diva2:547531
Bachelor of Science Programme in Computing Science