RoNS: Robust network function services in clouds

Abstract

In multi-tenant clouds, the traffic of tenants (e.g., enterprises) needs to be processed by network functions (NFs), for security and business logic issues. Due to potential hardware failures and software errors, NFs may break down. When encountering NF failures, we should consider two critical requirements for maintaining cloud robustness: limited influence scope and fast failure recovery. Without considering these two requirements, prior works based on deploying backup NF instances may result in large influence scope and long recovery time when a failure occurs. To bridge the gap, this paper investigates how to build robust network function services (RoNS) in multi-tenant clouds. Specifically, RoNS limits the number of tenants that each NF instance will serve so as to control the influence scope of an NF failure, and schedule requests with the help of agents designed in the data plane to achieve fast failure recovery. This is however a difficult undertaking. To solve this problem, RoNS takes a two-phase approach: NF instance allocation and tenant request scheduling. For NF instance allocation, we propose an efficient algorithm with bounded approximation factors based on the randomized rounding method. For tenant request scheduling, we present a primal–dual-based algorithm with a superior competitiveness ratio to solve it. We implement RoNS on a real testbed for experimental studies and use simulations for large-scale investigation. Both experiment results and simulation results show the superior performance of the proposed algorithms compared with other alternatives. For example, RoNS can cut down the number of affected tenants by 60%, and reduce recovery delay from 1170 ms to 316 ms on average, compared with existing failure recovery mechanisms based on deploying backup instances.