Cost-Minimal Reliability Allocation and Robustness of an Edge–Cloud Network Using the Coati and Arithmetic Optimization Algorithms

Authors

  • Hatem Kareem Sulaiman General Directorate of Education in Babylon, Ministry of Education, Iraq
  • Karrar Habeeb Hashim General Directorate of Education in Babylon, Ministry of Education, Iraq

Abstract

The cost of meeting a reliability requirement is rarely visible in weighted-objective allocation studies, and the robustness of the resulting design is almost never examined. This paper poses reliability allocation on a twelve-node heterogeneous edge–cloud
network (QSDR-ECRN) as cost minimization under an explicit reliability constraint, evaluated through the exact 69-term reliability polynomial of the network, which equals 0.998100 at the as-deployed component reliabilities. Two recent algorithms that have not been applied to network reliability allocation, the Coati Optimization Algorithm (COA) and the Arithmetic Optimization Algorithm (AOA), were implemented in Python and compared over 30 independent runs for each of five reliability targets from 0.9990 to 0.99995, with identical budgets of about 12,000 fitness evaluations, and against a constrained gradient-based reference solution. The cost-minimal designs form a smooth frontier in which only the storage node (R7) and the database node (R12) are upgraded, as exact Birnbaum importance suggests. COA stayed close to the reference optimum, with
a mean cost gap of 1.4% at the 0.9997 target and between 0.16% and 2.78% over all targets, whereas AOA settled consistently above it (12.6% at 0.9997, 3.4% to 16.0% overall); the difference was significant at every target (Wilcoxon rank-sum p < 0.001)
and for every budget and population size tested. A design optimized precisely for 0.9997 never meets that requirement in practice, a design targeted at 0.9998 meets it in only 1.5% of draws, and a design targeted at 0.9999 meets it in 99.4% of draws for 36.7% more cost, according to a Monte Carlo robustness study where the failure probability of every component rises by up to 50%. These numbers provide designers with a measurable cost for robustness.

Downloads

Published

2026-11-17