Dynamic load balancing in parallel and distributed networks by random matchings (extended abstract)
SPAA '94 Proceedings of the sixth annual ACM symposium on Parallel algorithms and architectures
Tight analyses of two local load balancing algorithms
STOC '95 Proceedings of the twenty-seventh annual ACM symposium on Theory of computing
Proceedings of the eighth annual ACM symposium on Parallel algorithms and architectures
An adversarial model for distributed dynamic load balancing
Proceedings of the tenth annual ACM symposium on Parallel algorithms and architectures
Hi-index | 0.00 |
We derive analytical results for a dynamic load balancing algorithm modeled by the heat equation ut = V2u. The model is appropriate for quickly diffusing disturbances in a local region of a computational domain without affecting other parts of the domain. The algorithm is useful for problems in computational fluid dynamics which involve moving boundaries and adaptive grids implemented on mesh connected multicomputers. The algorithm preserves task locality and uses only local communication. Resulting load distributions approximate time asymptotic solutions of the heat equation. As a consequence it is possible to predict both the rate of convergence and the quality of the final load distribution. These predictions suggest that a typical imbalance on a multicomputer with over a million processors can be reduced by one order of magnitude after 105 arithmetic operations at each processor. For large n the time complexity to reduce the expected imbalance is effectively independent of n.