SherlockFog: a new tool to support application analysis in Fog and Edge computing

Abstract

The Fog and Edge Computing paradigms have emerged as a solution to limitations of the Cloud Computing model to serve a huge amount of connected devices efficiently. These devices have unused computing power that could be exploited to execute parallel applications. A large number of existing and new parallel applications are programmed using Message Passing Interface, which is a de facto standard in High Performance Computing environments. We focus on the following question: Can MPI-based application take advantage of the increasing number of distributed resources available through Fog/Edge Computing Paradigm? In this work we present an extension to SherlockFog, a tool to experiment with parallel applications in Fog and Edge Computing environments to explore the impact of heterogeneity in computing power. This new version of our tool makes use of the Intel Pin Tool to inject instructions parametrically in the target code, mimicking different CPU computer power. A validation is presented using the MPI version of the MG and CG NAS Parallel Benchmarks to evaluate this estimation when used with SherlockFog to emulate Fog/Edge scenarios. We analyze the impact of slower nodes on two benchmarks and show that the incidence of a single slower node is significant, but slowing more nodes down does not further degrade performance. The latency effect is also analyzed, but its impact depends on the communication pattern of the target code. We show that SherlockFog provides a framework to analyze behavior of MPI libraries and applications towards achieving a Fog/Edge-ready distributed computing environment.