Mentor Amanda Bienz Mentee Yan Han CS Illinois
Mentor: Amanda Bienz Mentee: Yan Han CS @ Illinois Mapping a Partitioned Graph onto Processors Motivation • A computation that is allocated among a system of processors runs more efficiently since work is being done in parallel; however, communication between processors is slow. • We want to find the optimal distribution of a computation across a system of linked processors. • This problem can be modeled as the mapping of one graph onto another. The Processor Graph The Partitioned Graph • The second graph represents a system of processors, where each edge represents the delay between connected processors. • However, processors can communicate indirectly through each other, so we use the Floyd-Warshall algorithm to find the shortest delay between all processors. Average mapping score • The first graph represents the parts of the algorithm we want to map to the processors. • We divide the graph into partitions, creating a new graph where each node is a partition and edge weights equal the amount of connections between partitions. • The number of partitions should match the number of processors. 4000 Average mapping scores for different algorithms 3500 Random Algorithm A Algorithm B Best 3000 2500 2000 1500 1000 500 0 0 5 10 15 20 Number of vertices 25 30 35
- Slides: 1