How Far Can We Go in Compute-less Networking: Computation Correctness and Accuracy

Boubakr Nour,Soumaya Cherkaoui

DOI: https://doi.org/10.48550/arXiv.2103.15924

2021-12-09

Abstract:Emerging applications such as augmented reality and tactile Internet are compute-intensive and latency-sensitive, which hampers their running in constrained end devices alone or in the distant cloud. The stringent requirements of such application drove to the realization of Edge computing in which computation is offloaded near to users. Compute-less networking is an extension of edge computing that aims at reducing computation and abridging communication by adopting in-network computing and computation reuse. Computation reuse aims to cache the result of computations and use them to perform similar tasks in the future and, therefore, avoid redundant calculations and optimize the use of resources. In this paper, we focus on the correctness of the final output produced by computation reuse. Since the input might not be identical but similar, the reuse of previous computation raises questions about the accuracy of the final results. To this end, we implement a proof of concept to study and gauge the effectiveness and efficiency of computation reuse. We are able to reduce task completion time by up to 80% while ensuring high correctness. We further discuss open challenges and highlight future research directions.

Networking and Internet Architecture