Abstract: Metrics, including cost and latency, corresponding to a public computing system are monitored. A determination to migrate a workload running at donor computing system may be made based on the monitored metrics. A learning model, which may be initialized based on factors from a migration rules engine function, may make recommendations to migrate a workload based on a training data corpus. The learning model may be trained based on manual selections, or acceptances, of recommendations Automatic migration of a workload, under control of the learning model, may be made if recommendations are selected or accepted more frequently than an automatic migration criterion. Different learning models may be used to determine to migrate different corresponding workloads.

Abstract: Metrics, including cost and latency, corresponding to public computing system are monitored. A determination to migrate a workload running at a donor computing system may be made based on the monitored metrics. Before migrating the workload, state information corresponding to the workload may be stored to a remote storage that stores data used by the workload, but that is not part of the public computing systems. The recipient computing system may be instructed to retrieve the state information from the remote storage and to revive the workload based on the retrieved state information. The state information may comprise parameter values and factors that were current when the state information was stored. The state information may comprise an image of the workload. The recipient system may be instructed to revive the workload based on the retrieved state information retrieved from the remote storage.