Scheduling Many-Task Applications on Multi-clouds and Hybrid Clouds

A centralized scheduler can become a bottleneck for placing the tasks of many-task application on heterogeneous cloud resources. We have previously demonstrated that decentralized vector scheduling approach based performance measurements be used successfully this task placement scenario. then extended to latency measurements. Each node collects from its neighbors an overlay graph, measures communication latency, and makes local decisions where move tasks. Our recent experiments in CloudLab with nodes allocated multiple sites demonstrate using our results better resource utilization. While algorithm configuring graph was beneficial simulated delays, it not multi-cloud environment.