An Efficient Hybrid Scheduling Framework for Optimal Workload Execution in Federated Clouds to Maintain Performance SLAs

A Federated cloud is a composition of several clouds where a single federated cloud manager (FCM) is responsible for communication with the cloud service providers (CSPs) of associated clouds to accomplish the task of resource management. Finding optimal schedules for the incoming workloads from users is a highly complex task, as these workloads are expected to be completed under different deadlines. The existing frameworks for workload scheduling in federated cloud faced serious drawbacks such as inefficiency, unable to meet the deadlines assigned by users, etc. To overcome such drawbacks, this work introduces a new and effective strategy based on an optimization algorithm to achieve optimal scheduling of workloads. The proposed architecture involves a single FCM to identify the load in different clouds through communication with CSPs. A load level calculator (LLC) and a workload partitioning module (WPM) are maintained by the FCM to analyze the load in each cloud. Further, based on the load factor (LF) computed, the incoming workloads are partitioned into sub-queues of different sizes and are forwarded to the main clouds. The respective CSPs of the clouds maintain a cloud workload queue (CWQ) to locate the workloads and analyze the resource requirements. The CSP executes a scheduler based on hybrid flow-directed whale optimization (HFDWO) to find the optimal VMs in the data centre (DC) that can run the incoming workloads in the queue. The workloads are scheduled accordingly, and evaluations are conducted through simulations in the CloudSim tool. The performance of the approach is analyzed using the GWA T-12 Bitbrains dataset under different metrics. The overall improvement attained by the proposed approach compared to the existing frameworks is 25% in terms of SLA violation rate, 12% in terms of execution cost, 11% in terms of resource utilization, 33% in terms of makespan, 19% in terms of throughput and 28% in terms of response time.