Assessing Power Usage Effectiveness in Serverless Computing Environments

While computational resources continue to evolve and cloud computing continues to push boundaries, environmental concerns are increasingly questioning the cost of a broad spectrum of modern technologies. Among others, serverless computing has gained popularity in recent years due to its fast development, quick response time, dynamic resource allocation, auto-scaling, and fine-grained resource management capabilities. By default, serverless computing aims to save resources and energy, promoting greener computation. In this paper, we explore the Power Usage Effectiveness (PUE) of serverless computing by deploying and executing real experiments on a baremetal serverless platform equipped with the PowerAPI tool, to report on the energy consumption of containerized applications. We focus on understanding the virtual PUE (vPUE) and the combined PUE (cPUE) of serverless functions and platforms, considering the energy consumption of isolated functions, batches of functions, serverless platforms, and the application stack. We compute the energy profile of serverless computing benchmarks in different cluster settings and show the performance of executions as reserved resources increase on the same serverless platforms. Our studies demonstrate that a) pre-deployed functions achieve close to optimal vPUE and cPUE; b) not pre-deployed functions can achieve the same performance if used for longer executions; and c) reducing the number of available resources by a certain percentage while using a certain percentage of the remaining resources can significantly reduce the energy consumption of a serverless platform.