In-network Computing Based Compute Offloading Architecture For The Internet Of Things

Applications of the Internet of Things(IoT)have become part of our daily life during the last few years.Current Mobile Edge Compute(MEC)based architecture cannot meet the growing high-throughput computing(HTC)requirements as the number of IoT devices grows rapidly.Besides,MEC-based architecture suffers inherent uncertain delay and jitter.In recent,with the development of the programmable network hardware,the innetwork computing(INC)paradigm has received a large amount of attention.Benefited from the line-rate processing capacity and flexible programmability,INC presents a promising processing capacity in terms of high-throughput and low latency.In this paper,we introduce the INC paradigm into the IoT system.We design an INC based compute offloading architecture,named INMEC for IoT.The INMEC leverage computing resource inside the network to provide low-latency and high-throughput compute offloading service.To achieve both high capacity and versatility,we offload the critical subtasks to the INC devices while leave the rest to the MEC.Besides,we design an INMEC protocol for cooperation between INC systems and MEC systems.We also design a standard developing framework for data plane coding in P4 to support the INMEC.In the end,we present two use cases,Linear Quadratic Regulator(LQR)control and threshold control,to evaluate the feasibility of our architecture.And the experiment result have shown great improvement in terms of latency and bandwidth utilization.