An Efficient Flows Dispatching Scheme for Tardiness Minimization of Data-Intensive Applications in Heterogeneous Systems

Next-generation computer networks are expected to support a wide plethora of novel applications, demanding to handle a huge number of different computation tasks, typically data-intensive. As a consequence, the tasks raw data are pre-processed on Intermediate Processing Nodes (IPNs) before being sent toward the destination server. However, despite the IPNs are constrained in pre-processing capability, usually their influence on the resulting flow computing models is not taken into account. Aiming to fill this gap, this paper applies matching theory in defining a suitable flows dispatching scheme with the aim at maximizing the system utility in a capacity-constrained network. As a consequence, a novel matching algorithm with externalities is proposed, and the two-sided exchange stability is theoretically analyzed and properly discussed. We proved that the considered problem is NP-hard to solve and we verified the convergence of the proposed algorithm to a stable matching outcome, usually representing a severe challenge in the class of the matching games with externalities, due to the mutual dependence among the preferences lists of the players. Finally, simulation results evidence the validity of the proposed algorithm, highlighting remarkable advantages in comparison to different alternative schemes.