Self-adaptive Resource Allocation for Continuous Task Offloading in Pervasive Computing

Abstract

Task offloading has proven its potential in pervasive environments in numerous systems. In particular, code offloading has gained popularity as it allows to spontaneously forward work packages to remote resources. While for discrete tasks there are multiple systems that allow for code offloading already, stream processing has gained less research attention. In this paper, we propose a self-adaptive resource allocation approach for continuous task offloading. First, we tackle the issue of communication overhead by predicting future workload. We minimize not only the number of resource requests but also the scheduling delay. Second, we introduce a learning-based resource allocation mechanism that matches jobs and resource providers. The goal of the allocation mechanism is to assign jobs only to those resources that can finish a job in time. We use a code profiler to analyze the complexity of algorithms and perform machine learning to assign jobs to resources. Our results show that we can reduce the total communication overhead by more than 90 percent and assign jobs successfully with an F-Measure of .863.