TAIR: Achieving Tenant Anomaly Isolation with Request Scheduling in Serverless Computing

This chapter delves into the complexities of serverless computing, focusing on the critical aspects of load balancing and tenant isolation. It highlights the vulnerabilities inherent in multi-tenant serverless architectures, where weak inter-tenant isolation and untrusted function requests can lead to widespread service disruptions. The text introduces TAIR, a novel scheduling framework designed to address these challenges. TAIR employs a principle of Strong Isolation, ensuring that the set of nodes assigned to one tenant is never a subset of the nodes assigned to another, thereby preventing cascading failures. The framework uses a fast, submodular optimization-based algorithm to dynamically compute near-optimal scheduling plans in polynomial time. The chapter also provides a comprehensive evaluation of TAIR through extensive simulations and a system prototype, demonstrating its ability to prevent anomaly propagation with minimal performance overhead. By formalizing the problem of joint anomaly isolation and load balancing, designing an efficient approximation algorithm, and evaluating its performance, this text offers a robust solution to the pressing issues in serverless computing.