Revisiting the Performance of Serverless Computing: An Analysis of Variance

Serverless computing is an emerging cloud computing paradigm, which allows software engineers to develop applications at the granularity of function (called serverless functions). However, multiple identical runs of the same serverless functions can show different performance (i.e., response latencies) due to the highly dynamic underlying environment where these functions are executed. We conduct the first work to study serverless function performance to raise awareness of this variance among researchers. We investigate 59 related research papers published in top-tier conferences, and observe that only 40.68% of them use multiple runs to quantify the variance of serverless function performance. Then we extract 65 serverless functions used in these papers and find that the performance of these serverless functions can differ by up to 338.76% (44.15% on average), indicating a large magnitude of the variance. Furthermore, we find that 61.54% of these functions can have unreliable performance results at the low number of repetitions that are widely adopted in the serverless computing literature.