Apparatus and method for altruistic scheduling based on reinforcement learning
View Patent ↗The present disclosure relates to an apparatus and method of altruistic scheduling based on reinforcement learning. An altruistic scheduling apparatus according to an embodiment of the present disclosure includes: an external scheduling agent for determining a basic resource share for each process based on information of a resource management system; an internal scheduling agent for determining a basic resource allocation schedule for each process based on information including the basic resource share and a resource leftover based on the basic resource allocation schedule; and a leftover scheduling agent for determining a leftover resource allocation schedule based on information including the resource leftover. According to an embodiment of the present disclosure, it may be expected that reinforcement learning will not only mitigate the diminution of fairness of an altruistic scheduler but also further improve other performance indicators such as completion time and efficiency.
1. An altruistic scheduling apparatus based on reinforcement learning, the apparatus comprising:
one or more processors;
an external scheduling agent, executed by the one or more processors, for determining a basic resource share for each process based on information of a resource management system;
an internal scheduling agent, executed by the one or more processors, for determining a basic resource allocation schedule for each process based on information comprising the basic resource share and a resource leftover based on the basic resource allocation schedule; and
a leftover scheduling agent, executed by the one or more processors, for determining a leftover resource allocation schedule based on information comprising the resource leftover,
wherein the internal scheduling agent performs learning by receiving feedback from the resource management system reflecting at least one of the basic resource allocation schedule and the leftover resource allocation schedule, and
wherein the feedback comprises:
fairness evaluation on whether or not an opportunity to use a resource is equally provided to processes, evaluation of completion time that is a required time from the onset of a process to the end of execution, and efficiency evaluation based on the number of processes that are processed per unit time,
wherein the leftover scheduling agent performs learning by receiving feedback from the resource management system reflecting the basic resource allocation schedule or the leftover resource allocation schedule, and
wherein the feedback comprises the evaluation of completion time and the efficiency evaluation,
wherein the external scheduling agent performs learning by receiving feedback, and wherein the feedback comprises the fairness evaluation,
wherein the basic resource share comprises, for each process, information on a type of resource to be allocated, resource allocation time and a resource share,
wherein the basic resource allocation schedule comprises, for each task of each process, information on a type of resource to be allocated, resource allocation time and a resource share, and
wherein the information of the resource management system comprises:
a process request status comprising a set of newly requested processes and information associated with the newly requested processes; and
a resource usage status that is information on a resource needed by each process.
2. The apparatus of claim 1 , wherein the internal scheduling agent updates the resource usage status and the process request status by reflecting the basic resource allocation schedule in the resource usage status and the process request status respectively.
3. The apparatus of claim 2 , wherein the updated resource usage status and the updated process request status are comprised in information comprising the resource leftover.
4. An altruistic scheduling method based on reinforcement learning using external, internal and leftover scheduling agents, the method comprising:
determining, by using the external scheduling agent, a basic resource share for each process based on information of a resource management system;
determining, by using the internal scheduling agent, a basic resource allocation schedule for each process based on information comprising the basic resource share and a resource leftover based on the basic resource allocation schedule; and
determining, by using the leftover scheduling agent, a leftover resource allocation schedule based on information comprising the resource leftover,
wherein reinforcement learning using the internal scheduling agent is performed by receiving feedback from the resource management system reflecting the basic resource allocation system or the leftover resource allocation schedule, and
wherein the feedback comprises:
fairness evaluation on whether or not an opportunity to use a resource is equally provided to processes, evaluation of completion time that is a required time from the onset of a process to the end of execution, and
efficiency evaluation based on the number of processes that are processed per unit time,
wherein reinforcement learning using the leftover scheduling agent is performed by receiving feedback from the resource management system reflecting the basic resource allocation schedule or the leftover resource allocation schedule, and
wherein the feedback comprises the evaluation of completion time and the efficiency evaluation,
wherein reinforcement learning using the external scheduling agent is performed by receiving feedback, and
wherein the feedback comprises the fairness evaluation,
wherein the basic resource share comprises, for each process, information on a type of resource to be allocated, resource allocation time and a resource share,
wherein the basic resource allocation schedule comprises, for each task of each process, information on a type of resource to be allocated, resource allocation time and a resource share,
wherein the leftover resource allocation schedule comprises, for each task of each process, information on a type of leftover resource to be allocated, leftover resource allocation time and a leftover resource share,
wherein the information of the resource management system comprises:
a process request status comprising a set of newly requested processes and information associated with the newly requested processes; and
a resource usage status that is information on a resource needed by each process.
5. The method of claim 4 , wherein the resource usage status and the process request status are updated by reflecting the basic resource allocation schedule in the resource usage status and the process request status respectively.
6. A computer program, which is stored in a non-transitory computer-readable storage medium in a computer, for altruistic scheduling method based on reinforcement learning using external, internal and leftover scheduling agents, the computer program, executed by a processor of the computer, to perform:
determining, by using the external scheduling agent, a basic resource share for each process based on information of a resource management system;
determining, by using the internal scheduling agent, a basic resource allocation schedule for each process based on information comprising the basic resource share and a resource leftover based on the basic resource allocation schedule; and
determining, by using the leftover scheduling agent, a leftover resource allocation schedule based on information comprising the resource leftover,
wherein the internal scheduling agent performs learning by receiving feedback from the resource management system reflecting at least one of the basic resource allocation schedule and the leftover resource allocation schedule, and
wherein the feedback comprises:
fairness evaluation on whether or not an opportunity to use a resource is equally provided to processes, evaluation of completion time that is a required time from the onset of a process to the end of execution, and
efficiency evaluation based on the number of processes that are processed per unit time,
wherein the leftover scheduling agent performs learning by receiving feedback from the resource management system reflecting the basic resource allocation schedule or the leftover resource allocation schedule, and
wherein the feedback comprises the evaluation of completion time and the efficiency evaluation,
wherein the external scheduling agent performs learning by receiving feedback, and wherein the feedback comprises the fairness evaluation,
wherein the basic resource share comprises, for each process, information on a type of resource to be allocated, resource allocation time and a resource share,
wherein the basic resource allocation schedule comprises, for each task of each process, information on a type of resource to be allocated, resource allocation time and a resource share, and
wherein the information of the resource management system comprises:
a process request status comprising a set of newly requested processes and information associated with the newly requested processes; and
a resource usage status that is information on a resource needed by each process.