Improved Load Balancing in Large Scale Systems Using Attained Service Time Reporting

Our interest lies in load balancing jobs large scale systems consisting of multiple dispatchers and FCFS servers. In the absence any information on job sizes, a popular method is SQ( $d$ ) policy, which uses queue length reported by servers to assign incoming jobs. When sizes are highly variable, using only clearly suboptimal performance can be improved if some indication provided dispatcher about size an ongoing job. server measuring attained service time easy therefore report this together with when queried dispatcher. paper, we propose analyse variety policies improve policy exploiting both jobs, as well for used. We present unified analysis all these system under usual asymptotic independence assumptions. The accuracy proposed illustrated simulation. extensive numerical experiments indicate that significant improvement waiting (and thus also response) may achieved top server. Moreover, do not make use still provide moderately loaded systems.