Abstract: A computer-implemented method for computing workload management includes collecting power usage information from a location and collecting computing task information with respect to computing tasks performed at the location. The collected computing task information can be parsed to get a predicted approximate power usage information for each task. The tasks can be scheduled based on the predicted approximate power usage information to improve (e.g., maximize) usage of power from an unstable power source.

Abstract: Disclosed embodiments provide techniques for dynamic data collection. The dynamic data collection includes determining a data generation temporal pattern. Based on the data generation temporal pattern, a data collection strategy is created. The data collection strategy can be based on one or more data collection goals. The data collection strategy can contain specific details on how data is to be collected. A data infrastructure evaluation is performed, which provides pricing models for resources such as electricity and/or network bandwidth. A data collection policy is created based on the data collection strategy and the data infrastructure evaluation. The data collection policy can contain specific details on when data is to be collected and what strategy to use for the collection. A data transfer schedule is created based on the data collection policy. The data transfer schedule determines when to collect data from one or more data source devices.

Abstract: Provided are methods, systems, computer program products for scheduling jobs. The method may include receiving a request for allocating resources for a first job, a job comprising information regarding maximum amount of resources required by the job; determining a type of the first job; obtaining at least one backfill of the first job based on the determined type; allocating the maximum amount of resources from system resources to the first job; searching a second job in waiting jobs to be allocated resources, the second job being suitable to be allocated used resources by the first job from the maximum amount of resources allocated to the first job during the at least one backfill; allocating the resources unused by the first job from the maximum amount of resources allocated to the first job to the second job in response to the first job running to the at least one backfill.

Abstract: A processor may receive a request for a new computing task to be scheduled. The processor may identify that the new computing task requires processing in a cloud computing environment. The processor may identify resource needs of the new computing task. The processor may predict, utilizing an AI model, a future computing task based on the new computing task, where the future computing task requires processing at a future time in the cloud computing environment. The processor may identify resource needs of the future computing task. The processor may select one or more computing node hosts for processing the new computing task based on the future computing task.

Abstract: Provided are methods, systems, computer program products for scheduling jobs. The method may include receiving a request for allocating resources for a first job, a job comprising information regarding maximum amount of resources required by the job; determining a type of the first job; obtaining at least one backfill of the first job based on the determined type; allocating the maximum amount of resources from system resources to the first job; searching a second job in waiting jobs to be allocated resources, the second job being suitable to be allocated used resources by the first job from the maximum amount of resources allocated to the first job during the at least one backfill; allocating the resources unused by the first job from the maximum amount of resources allocated to the first job to the second job in response to the first job running to the at least one backfill.