The design and implementation of a CORBA load distribution servicefor distributed scientific computing applications running in a networkof workstations is described. The proposed approach is based onintegrating load distribution into the CORBA naming service which inturn relies on information provided by the underlying WINNER resourcemanagement system developed for typical networked Unix workstationenvironments. The necessary extensions to the naming service, the WINNERfeatures for collecting load information and the placement decisions aredescribed. A prototypical implementation of the complete system ispresented, and performance results obtained for the paralleloptimization of a mathematical test function are discussed
展开▼
