Determining instances to maintain on at least one cloud responsive to an evaluation of performance characteristics

摘要

A method includes the step of running a set of instances on at least one cloud for a first time interval, each of the instances comprising a bundle of virtualized resources. The method also includes the step of evaluating one or more performance characteristics of each of the instances in the set of instances over the first time interval. The method further includes the step of determining a first subset of the set of instances to maintain for a second time interval and a second subset of the set of instances to terminate for the second time interval responsive to the evaluating step. The steps are performed by at least one processing device comprising a processor coupled to a memory.

主权项

1. A method comprising the steps of:
running a set of instances on at least one cloud for a first time interval for a given processing job, each of the instances in the set of instances comprising a bundle of virtualized resources; evaluating one or more performance characteristics of each of the instances in the set of instances over the first time interval; separating the set of instances into a first subset of the set of instances to maintain for a second time interval and a second subset of the set of instances to terminate for the second time interval responsive to the evaluating step; starting at least one additional instance on the at least one cloud for the second time interval to replace at least one of the instances in the second subset; and migrating at least one of the instances in the second subset to the at least one additional instance; wherein evaluating the one or more performance characteristics comprises determining an average performance level of the set of instances over the first time interval; wherein separating the set of instances into the first subset and the second subset comprises:
determining a migration cost for migrating a given instance in the set of instances, the migration cost being amortized across an expected amount of time remaining for completion of the given processing job; andplacing the given instance in the second subset based at least in part on determining that a performance level of the given instance, taking into account the migration cost, is below the average performance level of the set of instances over the first time interval; and wherein the steps are performed by at least one processing device comprising a processor coupled to a memory.