An Experimental Assessment of Resource Queue Lengths as Load Indices
Technical Report Identifier: CSD-86-298
Abstract: Load indices that accurately reflect the current loads at computer system resources are crucial to the success of any dynamic load balancing scheme. However, few load indices have been experimentally validated as being suitable for load balancing. We conduct such a validation study for the resource queue lengths. We find that the CPU and disk queue lengths have very high correlation to the response times of CPU and I/O bound jobs, respectively. However, for the type of system that we studied, the system load changes very rapidly, making the response time high unpredictable. Simulation results suggest that load balancing will drastically reduce load fluctuation. The CPU is by far the most heavily used resource in the systems we studied. While this is also true in many other environments, measurement studies are called for before reaching such a conclusion in other systems.