Spatio-Temporal Resource Meshes for Serverless Computing

Hailiang Zhao,Shuiguang Deng,Jianwei Yin,Schahram Dustdar,Albert Y. Zomaya

2021-01-01

Abstract:—Serverless computing is leading the way to a simplified and general purpose programming model for the cloud. A key enabler behind serverless computing is efficient load balancing, which routes function invocations to appropriate backend VM nodes. However, current load balancing algorithms implemented in serverless platforms are static and simple to operate without advantageous performance guarantee. Although policies such as Pod or JFIQ yield asymptotically optimal average response time, the information they depend on are usually unavailable. In addition, dispatching serverless functions with strict deadlines to heterogenous VM nodes with limited input data processing speeds online is difficult because the resource contention is intensified with frequently creating and evicting of function instances. To design an online load balancing algorithm without assumptions on the arrival patterns of serverless functions and the processing capacity of VM nodes while maximizing the social welfare, i.e., the sum of utility of functions and the revenue of the serverless platform, we construct several pseudo-social welfare functions and marginal cost functions, where the latter is to estimate the marginal cost for provisioning resources to every newly arrived function based on present resource surplus. The proposed algorithm, named OnSocMax, works by following the solutions of several convex optimization problems. It is proved to be α -competitive for some α at least 2. We also validate the theoretical superiority of OnSocMax with simulations and the results show that it significantly outperforms two handcrafted benchmarking policies on the social welfare.