Refactoring Network Functions Modules to Reduce Latencies and Improve Fault Tolerance in NFV

摘要

Network functions virtualization (NFV) allows service providers to deliver new services to their customers more quickly by adopting software-centric network functions implementation over commercial, off-the-shelf hardwares. This NFV-based software-centric approach cannot use dedicated mechanisms implemented over custom built boxes to reduce latencies and tolerate faults. We present a case study of IP multimedia subsystem (IMS), which is the most complex NFV instance, requires extremely low end-to-end latency (40 msec), and demands system availability as high as five nines. Through an empirical study, we discover that highly modular IMS network functions implementation over virtualized platform: 1) incurs latencies and 2) does not tolerate faults. NFV-based IMS modules incur high latencies by creating a feedback loop among each other while executing delay sensitive data-plane traffic. These IMS modules are also susceptible to failure, causing the control-plane to terminate the application session while keeping the data-plane to forward data packets. To address these issues, we propose to refactor network function modules. We reduce latencies by pipelining the IMS modules, and recover failed modules by reconfiguring their neighboring modules. We build our system prototype of open source IMS over OpenStack platform. Our results show that our scheme reduces latencies and failure recovery time up to$12imes $and$10imes $, respectively, when compared with the state-of-the-art virtualized IMS implementation.