The effect of network delay estimation error on computation offloading decisions

摘要

We study a computation offloading system with multiple cloud resources that can serve as offload targets. In this system, given a computational job, an M-ary decision must be made to select a local or remote (offload) compute resource where the job will be executed. Such a decision is made using estimators that approximate the job's completion time for each of the M resources. The effectiveness of a decision and, thus, the selected resource, can be measured by the decision difference. To characterize the effect of the network delay estimation error on the decision difference, we develop and solve a stochastic model using the Monte Carlo simulation technique in a wide range of simulations. The simulation results indicate that the network delay estimation error has no impact on the decision difference until a certain inflection point, at which the error term begins to result in incorrect decisions. These findings provide insight into the appropriate selection of estimators that enable computation offloading decision-making.