Serverless Function Latency Model in Edge Computing
TL;DR
A statistical model of the service latency of serverless functions, with particular reference to edge computing, based on the observation of experimental latency values and adapted to produce a statistical distribution that closely approximates the real one is illustrated.
Abstract
This paper illustrates a statistical model of the service latency of serverless functions, with particular reference to edge computing. Given a set of functions deployed in an edge computing system, this model can be very useful for several reasons. First of all, evaluating the quality of the user experience is essential to promptly assess whether the system configuration state requires corrective interventions. In case of malfunctions, the model may highlight performance problems. The model can also identify possible Service Level Agreement violations in real time, before they can cause operational problems. In addition, it allows balancing cost-performance trade-offs. The proposed model is based on the observation of experimental latency values and is adapted to produce a statistical distribution that closely approximates the real one. Furthermore, it can be exploited synergistically with Machine Learning algorithms, to dynamically optimize serverless systems in edge computing. We have evaluated the performance of the model by using latency samples generated ad hoc through a mixture of distributions, including long tailed ones, and by using the well-known Azure Functions and Globus datasets. The experimental results show a significant closeness of the statistical distribution of the generated data with the experimental ones, up to high percentiles.