A distributed load‐balancing system and its application to list‐processing oriented data‐flow machine dfm
Masaru Takesue · Systems and Computers in Japan · 1988
Abstract This paper proposes a group load‐balancing system (GLB) for the multiprocessor system which can manage and allocate the load in parallel and in a distributed way. In GLB, the resources are grouped hierarchically and the request for the resource allocated to a group is assigned to the subgroup with the minimum load in that group. Consequently, the group load data are updated at each allocation time, maintaining the accurate value, leading to a high load‐balancing ability. GLB is applied to the data‐flow machine DFM II and the evaluation was made by RT level simulator. The following result was obtained: (1) The performance is improved by 20 to 30 percent compared with the random allocation; (2) the maximum performance is obtained when the processor load is set equal to the number of active state processes; (3) the load balancing can be achieved within the difference of 5 processes on the average.