A solution to QOS control and availability promotion in complex grid computing

Complex computational tasks are combinations of multiple independent computational subtasks related by data dependency. Grid environments bring out new challenges to their running, one of which is the normality and performance. To guarantee the running quality and survive failures, availability management and multiple quality promotion ways are adopted in the process of complex task running. It requires no or little change to legacy computational programs. Quantified user's requirements are the goal to performance measuring and availability management. In the process of a task running, the availability management framework monitors, analyzes performance measurements, controls performance degradations, and tolerates failures. Besides this, multiple auxiliary methods help improve the effectiveness of quality control, which includes: to construct computation-matched virtual environments, to raise the correctness of program running, and to decrease management costs. The grid computation running support system implemented under this design is extensible, adaptable, universal, stable and highly available. ? 2004 IEEE.EI