Convergence-Guided Iterative Optimization for Dynamic Task Scheduling in Distributed Computing Systems

Abstract

Distributed computing systems contain one of the most serious optimization problems in dynamic task scheduling; this operates under varying workloads, resource allocation and availability, communication latency, and Queue conditions.Several scheduling approaches mostly depend on immediate observations, limited historical data, and heuristic optimization. Theseapproaches do not adequately capture heterogeneous temporal dependencies occurring at different time periods. This paper describes a multi-timescale fractional memory framework for dynamic task scheduling. It shows short-, medium-, andlong-term historical effects in a weighted fractional derivative combination of different orders. Dynamic weights can help with the relative combination of all memory parameters according to the observed dynamics of the system. Thisscheduling method is executed as a constrained multi-term fractional optimization problem, and the resulting method structure is analyzed with respect to well posedness, convergence, disturbance-induced boundedness, and numerical implementation. The proposed formulation indicates memoryless and single-order fractional scheduling models by allowing multiple temporal memory mechanisms in order to influence resource allocation continuously. This work depicts a mathematical connection in-between multi-term fractional dynamical systems and adaptive scheduling in distributed computing environments.