Environment-Sensitive Performance Tuning for Distributed Service Orchestration

Modern distributed systems are designed to tolerate unreliable environments, i.e., they aim to provide services even when some failures happen in the underlying hardware or network. However, the impact of unreliable environments can be significant on the performance of the distributed systems, which should be considered when deploying the services. In this paper, we present an approach to optimize performance of the distributed systems under unreliable deployed environments, through searching for optimal configuration parameters. To simulate an unreliable environment, we inject several failures in the environment of a service application, such as a node crash in the cluster, network failures between nodes, resource contention in nodes, etc. Then, we use a search algorithm to find the optimal parameters automatically in the user-selected parameter space, under the unreliable environment we created. We have implemented our approach in a testing-based framework and applied it to several well-known distributed service systems.