The influence of different workload descriptions on a heuristic load balancing scheme

A task scheduler based on the concept of a stochastic learning automation, implemented on a network of Unix workstations, is described. Creating an artificial, executable workload, a number of experiments were conducted to determine the effect of different workload descriptions. These workload descriptions characterize the load at one host and determine whether a newly created task is to be executed locally or remotely. Six one-dimensional workload descriptors are examined. Two workload descriptions that are more complex are also considered. It is shown that the best single workload descriptor is the number of tasks in the run queue. The use of the worst workload descriptor, the 1-min load average, resulted in an increase of the mean response time of over 32%, compared to the best descriptor. The two best workload descriptors, the number of tasks in the run queue and the system call rate, are combined to measure a host's load. Experimental results indicate that no performance improvements over the scheduler versions using a one-dimensional workload descriptor can be obtained.<
>