Static process allocation using information about program behaviour
Lars Lundberg · 2002
Efficient allocation of processes to processors is important in multiprocessors. Some systems cannot tolerate the overhead enforced by process migration; such systems have to use static allocation In order to perform well static allocation strategies require some knowledge about the behaviour of the parallel program. If the behaviour of the parallel program does not change from one execution to another then recordings from previous executions could facilitate efficient allocation. In the paper a way of using recordings from previous executions is explored. A static allocation algorithm using such information is compared to a static algorithm lacking this information and to a dynamic algorithm. Measurements show that the allocation strategy using information about program behaviour yields a better speed-up than the algorithms lacking this information.>