Benchmarking OpenStack for edge computing applications

In this paper, we focus on the identification and evaluation of performance factors of OpenStack-based edge computing platforms. Such infrastructure relies on the deployment of additional computing resources close to the data source, to alleviate low throughput, latencies and network congestion. While cloud data centres offer numerous compute-intensive processing units, the edge layer leverages heterogeneous, dynamic and performance-constrained computing devices. The edge avoids the systematic use of low Quality of Service (QoS) of Wide Area Network (WAN) links to perform computations. Many edge platforms incorporate cloud software environment provisioning and elasticity capabilities. In this regard, we investigate the performance of the state-of-the-art OpenStack framework to set up an edge computing infrastructure. To achieve our goal, we first evaluate the impact of the deployment type, whether bare metal, containerised, virtualized or emulated on compute-intensive kernels. Within the same deployment scenarios, we examine the OpenStack virtual network infrastructure latency and throughput. We also cover the duration of several OpenStack primitives involved in the deployment of applications. Finally, we execute an MPI (Message Passing Interface) distributed application and highlight the performance impact of our infrastructure. Our study provides a comprehensive analysis of parameters influencing performance in an edge computing infrastructure based on OpenStack.