Optimization of mixed database workload scheduling and concurrency control by mining data dependency relationships via lock tracking

摘要

The present disclosure relates generally to the field of determining data access patterns associated with different workloads. In various examples, determining data access patterns associated with different workloads may be implemented in the form of systems, methods and/or algorithms. The present disclosure also relates generally to the field of scheduling workloads (and/or recommending scheduling for workloads) based upon the determined data access patterns associated with the different workloads. In various examples, scheduling workloads (and/or recommending scheduling for workloads) may be implemented in the form of systems, methods and/or algorithms.

主权项

1. A method for workload scheduling, the method comprising:
obtaining, by a processor, first lock trace data associated with a plurality of workloads, the first lock trace data relating to a first set of objects; replaying, by the processor, the first lock trace data associated with the workloads to identify first data access conflicts between a first one of the workloads and a second one of the workloads; determining by the processor, based upon the identified first data access conflicts associated with the first lock trace data, a subset of the first set of objects; obtaining, by the processor, second lock trace data associated with the workloads, the second lock trace data relating to the subset of objects; replaying, by the processor, the second lock trace data associated with the workloads to identify second data access conflicts between the first one of the workloads and the second one of the workloads; decomposing by the processor, based at least in part upon the identified second data access conflicts, at least the first one of the workloads into a plurality of shorter duration execution units; and scheduling, by the processor, at least one of the shorter duration execution units at a first time window that at least partially overlaps a second time window of at least the second one of the workloads.