91av

A Japanese team has increased the speed of parallel computing by using a
genetic algorithm to schedule processing tasks. Tatsuhiro Tsuchiya and
colleagues at Osaka University have designed their evolutionary algorithm to
allow only the “fittest” task sequences—the computer equivalents of
chromosomes—to survive (Microprocessors and Microsystems, vol 22,
p 197). The surviving sequences become the tasks executed by each processor node
in an array. Using task duplication, so that the task can be executed as soon as
a node is free, the algorithm performed as well as—and sometimes
better—than DSH, a common non-genetic task-scheduling system.