Lasp on Grisp : implementation and evaluation of a general purpose edge computing system for Internet of Things

Details

Supervisors

Van Roy, Peter

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en informatique
Master [120] en sciences informatiques

Abstract

The cumulus needs to become stratus. In other words, the current cloud computing in- frastructure needs to dissipate and relegate some of its workload to nodes that are closer to clients, at the edge of networks. This requirement is a direct consequence of the exponential growth of the amount of Internet of Things (IoT) devices. Even- though this need for a new edge computing paradigm is acknowledged by all, there is currently no implementation of it and IoT applications still resort to traditional designs. One of the biggest challenges for a production grade edge model is outperform- ing cloud datacenters for general purpose applications. Edge networks can imply both high heterogeneousness and unreliability, and nodes can be severely resource constrained. LightKone 1 is an European project that aims at solving this issue by bringing a novel extension to the cloud, where all levels starting from edge nodes coalesce their efforts to alleviate the burden of the layers above them. By combining synchronization-free programming and hybrid gossip algorithms, LightKone introduces a design for general purpose tasks at the edge. This model turns the 50 billions of additional IoT devices that are predicted for 2025 into sup- plementary resources for the cloud. They will be able to provide similar services as currently available cloud computers, but directly on the edge. In this context, we propose GrispLasp, a prototype implementation of an edge node that embodies LighKone’s vision of an edge device. The key objective of this work is to provide a concrete general purpose edge network, and evaluate its overall efficiency and performance. We have implemented a Lasp distributed application framework running on GRiSP boards. It is augmented with a task model that en- ables generic computations and distributed storage through Lasp. We have made real-life deployments under the forms of wireless sensor networks and were able to successfully demonstrate that IoT sensor data gathered from the GRiSP boards could be aggregated, processed and stored directly in the GrispLasp edge. Our prototype was thoroughly benchmarked against its counterpart, that would run identical applications but would rely on the services of Amazon EC2 instances for storage and processing. Our first results indicate that this new programming paradigm could alleviate the hardness of building applications on the edge. We expect that future edge nodes become more powerful and confirm our results, and increase scalability for generic edge computations.