Serverless Platforms Performance Evaluation at the Network Edge

Abstract

Emerging computer paradigms aim to fulfill the ever-present ideal of running as many applications on existing infrastructure, as efficiently as possible. One such novel concept is serverless computing which abstracts away infrastructure management, scaling and deployment from developers, allowing them to host function instances with granular responsibilities. However, faced with the meteoric growth in the number of IoT devices, the cloud is no longer suitable to meet this demand and a shift to edge infrastructures is needed, providing reduced latencies. While there are existing serverless platforms, both commercial and open-source that can be deployed at the edge, a comprehensive performance analysis is needed to determine their advantages and drawbacks, define open-issues, and identify areas for improvement. This paper analyses three different serverless edge platforms with the help of an existing serverless test suite, outlining their architecture, as well as execution performance in both sequential and parallel invocation scenarios. Special focus is paid to solutions that can be deployed in a standalone fashion, without complex clustering requirements. Results show that while the serial execution performance is comparable among the analyzed platforms, there are noticeable differences in cases of concurrent executions.