Amazon.com, Inc.
Preventing performance degradation over time in a cache

Abstract:

Systems and techniques are described for predicting future overlap in requests for a compute resource to address a system overload before it occurs. Requests for a resource may be tracked in time and grouped based on one or more common characteristics of the requests, such as a time of occurrence of the requests and a period that they repeat. Once grouped, different groups of requests for the resource may be tracked across at least one dimension, such as a periodic time of occurrence, a volume of requests, or a length of time of each occurrence of a group, to generate tracking data. Based on the tracking data, predictions may be generated indicating whether and to what extent the groups of resources will overlap at a future time. Additional resources may be provisioned to process the requests to prevent or reduce the likelihood of a system overload at the future time.