Jupiter: a networked computing architecture

Modern latency-sensitive applications such as real-time multi-camera video analytics require networked computing to meet the time constraints. We present Jupiter, an open-source networked computing system that inputs a Directed Acyclic Graph (DAG)-based computational task graph to efficiently distribute the tasks among a set of networked compute nodes and orchestrates the execution thereafter. This Kubernetes container-orchestration-based system includes a range of profilers: network profilers, resource profilers, and execution time profilers; to support both centralized and decentralized scheduling algorithms. While centralized scheduling algorithms with global knowledge have been popular among the grid/cloud computing community, we argue that a distributed scheduling approach is better suited for networked computing due to lower communication and computation overhead in the face of network dynamics. We propose a new class of distributed scheduling algorithms called WAVE and show that despite using more localized knowledge, the WAVE algorithm can match the performance of a well-known centralized scheduling algorithm called Heterogeneous Earliest Finish Time (HEFT). To this, we present a set of real-world experiments on two separate testbeds: (1) a worldwide network of 90 cloud computers across eight cities and (2) a cluster of 30 Raspberry pi nodes.