Calculating scheduler’s contamination rate

Resum

In distributed computing infrastructures(DCIs), schedulingalgorithms based on hardware energy efficiency measuresare not suitable because these algorithms need a completeknowledge of the whole system. Gathering all thisinformation about pending tasks and available resourcescharacteristics is impossible in decentralized architectureslike are DCIs. Instead, in this scenarios, we propose theutilization of performance based scheduling policies andresource sharing strategies. The main idea is toanticipate tothe resourcess saturation, becausethis means that less jobswill have to be rescheduled. In this way, these policies canreduce algorithm’s total communication’s cost. In addition,another problem of energy-aware schedulers is that they donot provide information about the overhead they introducein the system due to scheduling jobs and monitoring them.Thus, this paper introduces the concept ofschedulerscontamination rate. This measure indicates the grams ofCO2 created by the scheduler per job executed.Our resultsshow that a default scheduler produces almost three timesmore grams of CO2 per job executed than our anticipate toresourcess saturation scheduler. All in all, our schedulingmechanism is beneficial to everyone: it consumes lessenergy, big infrastructures owners can achieve their owninternal objectives and small users can reach the bestpossible completion time for their applications.