On the Modelling and Analysis of Edge-Serverless Computing

Abstract

The exponential increase in data generated by mobile devices and the rapid growth of cloud computing traffic are pushing services to the edge of the network. Serverless computing is an emerging cloud computing paradigm that provides stateless and event-driven services, called serverless services, independent of network infrastructure. Further improvements in quality of service (QoS) can be achieved by utilizing novel concepts like edge-serverless computing. Analyzing such systems as in existing works requires network-level simulations which could be costly and time-consuming, thus analytical models to obtain rapid insights are also needed. In this paper, we propose a stochastic geometry-based mathematical model in a tiered network, characterizing average end-to-end delay as performance measure, impacted by various crucial network parameters. The spatial distribution of server nodes, as well as computation requests are randomly and independently distributed according to homogeneous Poisson point processes (PPPs), allowing us to study the performance gains from multiple aspects. Two distinct network topologies are investigated numerically, namely communication and computation-oriented, validating the accuracy of our theoretical approximations.