Scheduling Many-Task Computing Applications for a Hybrid Cloud

A centralized scheduler can become a bottleneck for placing the tasks of a many-task application on heterogeneous cloud resources. Previously, it was demonstrated that a decentralized vector scheduling approach based on performance measurements can be used successfully for this task placement scenario. In this dissertation, we extend this approach to task placement based on latency measurements. Each node collects performance metrics from its neighbors on an overlay graph, measures the communication latency, and then makes local decisions on where to move tasks. We present a decentralized and a centralized algorithm for configuring the overlay graph based on latency measurements and extend the vector scheduling approach to take latency into consideration. Our experiments in CloudLab, both in a simulated environment and in realistic conditions, demonstrate that this approach results in better performance and resource utilization than without latency information