In order to examine some unfair assignment decisions the VMS scheduler makes under certain load conditions, this session deals with the interaction between the processes present, in function of their resource needs, focusing on the assignment distribution of the CPU between the processes. A performance index is defined to outline the service rate of the single processes, which is not dependent on their characteristics, in order to be able to compare processes of different requirements. A discussion is presented on the different ways of calculating this index, outlining the difficulties incurred into for the practical retreivement of the data. The actual measure is done by a specially written monitor capable of retreaving data at run time, and then examining it, in order to keep the influence of the monitor to a minimum. Extensive studies have been made to understand ami minimize the influence that the presence of the monitor has on the scheduler's behaviour. Various criticai load distributions are examined, ami a feedback method for correcting these problems has been delineated. This acts on the preemption round robin scheduling method of the VMS 05, adding to the Pixscan utility already present.
Measuring and Enhancing Scheduling under VMS
Lancia M;
1994
Abstract
In order to examine some unfair assignment decisions the VMS scheduler makes under certain load conditions, this session deals with the interaction between the processes present, in function of their resource needs, focusing on the assignment distribution of the CPU between the processes. A performance index is defined to outline the service rate of the single processes, which is not dependent on their characteristics, in order to be able to compare processes of different requirements. A discussion is presented on the different ways of calculating this index, outlining the difficulties incurred into for the practical retreivement of the data. The actual measure is done by a specially written monitor capable of retreaving data at run time, and then examining it, in order to keep the influence of the monitor to a minimum. Extensive studies have been made to understand ami minimize the influence that the presence of the monitor has on the scheduler's behaviour. Various criticai load distributions are examined, ami a feedback method for correcting these problems has been delineated. This acts on the preemption round robin scheduling method of the VMS 05, adding to the Pixscan utility already present.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


