Multicore processors are becoming more and more widespread, specially in the server market. Data centers can harvest the power of multicore systems once proper scheduling methods are comprehended. Current operating systems do not consider all characteristics of the applications that are being scheduled and therefore cannot make optimal scheduling decisions. This will waste the power of multicore systems and increase the costs of a data center. The idea explained in this thesis is to solve one of the problems of scheduling on multicore systems. Using the methods introduced in this thesis, operating systems can detect data sharing between dierent threads of a multithreaded application and make better scheduling decisions. Sharing aware scheduling can improve the performance of applications by up to 42%. The scheduler can detect data sharing dynamically just by monitoring hardware performance counters.
Copyright is held by the author.
The author granted permission for the file to be printed and for the text to be copied and pasted.
Supervisor or Senior Supervisor
Thesis advisor: Fedorova, Alexandra
Member of collection