چكيده به لاتين
As a combination of time-sharing and space-sharing, in Gang scheduling tasks are grouped together, and they are executed on processors, simultaneously. Although some processors are idle, gangs are inevitably queued. This approach makes fragmentation in the processing space. One solution technique to overcome this pitfall is using Migration, however it requires consideration of migration overheads. In order to make gang scheduling decisions, Ousterhout matrix with four main steps is employed. The second step that is called Compact matrix is of significant importance due to playing a key roll in scheduling gangs, thereby increasing scheduling speed in other gangs, while reducing the mean response time.
Improved Compact Matrix Algorithm (ICMA) utilizes two distinct concepts, namely Difference Criterion and Exchange Idea to improve the performance of the compact matrix. Moreover, it combines gaps for scheduling other gangs, reducing the mean response time. However, with the increase in the migration overhead, the mean response time also increases as compared to the Basic Scheduling with Migration (BasicwM), because the number of occuring migration is much higher than it. In this thesis, high migration overheads are effectively prevented by setting a predefined threshold level for the migration cost.
Simulation results achieved from executing the proposed Constrained Migration Gang scheduling (CMG) in different conditions, indicate that the mean response time as well as the weighted mean response time have been conspicuously reduced, compared to the (BASICwM) method. In this regard, it is observed for 20, 25, and 30 percent of the time slots, the mean response time have been declined to 9% to 15%, 6.1% to 12.7%, and 2.25% to 7.8 percent. The simulation results also prove the better performance of the proposed method for smaller migration overheads and the heavier workloads. Furthermore, it can be inferred from the simulation results that adoption of AFCFSwM policy induces a better performance as compared to LGFSwM policy.
