Division of load for operating system kernel

Abstract

The current Operating System (OS) kernels calculate the load average value as a lump sum. Also the algorithm for the calculation of load average does not separate CPU load from Disk load. This leads to the presentation of an incorrect measurement when both disk bound tasks and CPU bound tasks run simultaneously. In this paper a new algorithm is proposed to calculate, store and display each user's CPU and Disk loads separately. The separation of user load at the kernel level has an importance in the collection of historical load signals as they can be useful for load prediction. In Grids and Clusters the users have certain usage patterns that can be easily traced back in the historical load profile collections. Such selected patterns are useful in the prediction of load profiles.