Outcomes of on-line configuration
The actual complexity lies in an internet setting the place jobs arrive dynamically and the scheduler has to make irreversible choices on the fly with out figuring out which job will arrive subsequent. We quantified the efficiency of the net algorithm by the competitors ratio. This can be a worst-case comparability between the throughput of the net algorithm and the throughput of the optimum algorithm that is aware of all jobs a priori.
Commonplace non-preemptive algorithms fail completely right here, because the competitors fee approaches zero. This happens as a result of one unhealthy choice to schedule a protracted job can break the potential of scheduling many smaller jobs sooner or later. On this instance, whatever the size of the finished job, for those who think about that every accomplished job has equal weight, finishing many brief jobs is rather more worthwhile than finishing one lengthy job.
To make on-line issues solvable and replicate real-world flexibility, we studied two fashions wherein energetic jobs may be interrupted if a greater alternative arises (however solely jobs which might be restarted after which accomplished non-preemptively are counted as profitable).
Interruption resulting from reboot
On this mannequin, an internet algorithm permits you to interrupt a presently working job. Any partial work already accomplished on the interrupted job is misplaced, however the job itself stays on the system and may be retried.
We now have discovered the pliability supplied by permitting jobs to renew to be extraordinarily helpful. A variant of Grasping that repeatedly schedules the earliest-finishing job continues to realize a competition fee of 1/2, which is akin to the leads to the offline setting.
Interruption with out restart
On this extra strict mannequin, all work carried out in an interrupted job is misplaced, and the job itself is completely discarded. Sadly, this rigorous mannequin reveals that any on-line algorithm can encounter a set of jobs that drive choices that stop it from satisfying extra duties sooner or later. Once more, the competitors fee for all on-line algorithms is near zero. Our evaluation of the exhausting cases above led us to give attention to sensible eventualities the place all jobs share a typical deadline (for instance, all information processing have to be accomplished by the nightly batch run). For such widespread deadline instances, we devise a brand new fixed competitors algorithm. Our algorithm could be very intuitive, and right here we describe an algorithm that enables easy configuration of unit capability profiles, i.e., scheduling a single job at any time.
On this setting, the algorithm maintains a tentative schedule by assigning already arrived jobs to completely different time intervals. When a brand new job arrives, the algorithm modifications the tentative schedule by performing the primary of 4 actions:
