Power efficient distribution and execution of tasks upon hardware fault with multiple processors

摘要

Tasks may be scheduled on more than one processor to allow the processors to operate at lower processor frequencies and processor supply voltages. Multiple processors executing tasks in parallel at lower frequencies and supply voltages may allow completion of the tasks by deadlines at lower power consumption than a single processor executing all tasks at high frequencies and supply voltages. Power efficiency of a computer system may be improved by using a combination of processors executing tasks using a combination of earliest deadline first (EDF), earliest deadline last (EDL), and round robin (RR) queue management methods.

主权项

1. A method, comprising:
detecting, by a processor, at least one processor, scheduled to execute portions of a queue of realtime tasks and a queue of non-realtime tasks, has failed of a group of processors spanning at least two platforms coupled by a network; determining, by the processor, whether the failed processor of the group of processors is local to the processor or whether the failed processor of the group of processors is coupled through a network to the processor; and performing, by the processor, a course of action for performing tasks assigned to the failed process based, at least in part, on whether the failed processor is a local processor or a cloud processor; detecting, by the processor, that a second processor, scheduled to execute portions of the realtime tasks and non-realtime tasks, has failed of the group of processors; scheduling, by the processor after detecting the second failed processor, the realtime tasks on a first other processor of the group of processor; executing the realtime tasks on the first other processor using at least one execution strategy selected from a group comprising using reliability aware power management (RAPM) and using earliest deadline first (EDF) at a maximum frequency of the first other processor, the execution strategy selected based, at least in part, on a workload of the first other processor; when executing realtime tasks using reliability aware power management (RAPM), executing the non-realtime tasks in idle intervals between executing the realtime tasks; when executing realtime tasks using earliest deadline first (EDF), executing the non-realtime tasks in idle intervals between executing the realtime tasks.