Identifying and alleviating shared cache contention : Achieving reliability of real-time tasks on a multi-OS and multi-core system

Current architecture trends results in processors being equipped with more cores and larger shared caches. Concurrent applications on multicore processors may interfere with each other when accessing shared resources. This is especially troublesome if deadline-bound real-time tasks are running. A tool illustrating contention was developed. The tool was used to confirm the contention problem and to evaluate the developed solution. This thesis surveyed state-of-the-art approaches concerned with mitigating contention. The approaches can be categorized as requiring modifications to the operating system, requiring modifications to hardware, requiring both or requiring neither. The approaches were also characterized by whether they focused on the source of contention or the contended for resource. An approach involving throttling of individual cores by clock modulation and toggling of hardware prefetchers was developed and tested. The solution was demonstrably effective in reducing contention. Contention effects were not eliminated. Possible further work include improving autonomous detection of contention and accounting for, and illustrating contention effects involving, additional contended for resources