In this talk, we describe challenges in modeling and analyzing queueing systems with incomplete information, which is often the case arising from large-scaled systems with dependence such as various Internet applications involving big data. We try to use Bayesian approach for queueing models with miss information. We first start with a simple example to demonstrate basic ideas and concepts and then show how Bayesian approach works for the join-the-shortest-queue (JSQ for short) model with two parallel queues, where the servers may be heterogeneous. Since in practice, the complete data are not always (or seldom) available, we are dealing with the uncertainty of the JSQ parameters in case of missing information. The uncertainty is, indeed, caused by the lack of information (to the observer or customers) about the desired queueing model, resulted in situations where the shortest queue is not always assigned (by the system) to the upcoming jobs. We show how a Bayesian approach can be applied to estimate unknown system parameters and to how the prediction can help the system to assign the shortest queue to the upcoming jobs more accurately.
This talk is based on the joint work with Ehssan Ghashim.