Real-time network function virtualization with timing interfaces

More and more infrastructure is becoming virtualized. Recently this trend has begun to include network functions - such as firewalls, WAN optimizers, and intrusion prevention systems - that have traditionally been implemented as middleboxes using dedicated hardware. This trend towards network function virtualization (NFV) offers a variety of potential benefits that resemble those of cloud computing, including consolidation, easier management, higher efficiency, and better scalability. However, current cloud technology is not a perfect match for NFV workloads: since the infrastructure is shared, the time it takes for a packet to pass through a particular function is no longer predictable, and can in fact vary considerably. This is causing headaches for operators, who can no longer treat network functions as "bumps in the wire" and must now consider a complex web of possible interactions and cross-talk when operating or diagnosing their systems. In this position paper, we propose a compositional approach towards building a scalable NFV platform that can provide latency and throughput guarantees using timing interfaces. We discuss our preliminary results that leverage and extend recent advances on timing interfaces and compositional theory from the real-time systems domain to the NFV setting, and we highlight open challenges and potential directions towards real-time NFV.