Comparison of Optimization Techniques for Risk, Deploy-Cost, and Makespan Aware Network Synthesis for Dispersed Computing
Jared Coleman
Abstract:
For this poster presentation, we evaluate different optimization techniques for solving the RDM (Risk, Deploy-Cost, Makespan) Network Synthesis problem, a internet of battlefield things inspired network synthesis for dispersed computing problem originally proposed in the paper “Multi-Objective Network Synthesis for Dispersed Computing in Tactical Environments”.
In particular, we consider networks of compute nodes and communication links between them that are either high- or low-speed, high- or low-risk, and high- or low-cost (to deploy).
The goal is to find the subset of nodes which best supports a given distributed application, modeled as a directed acyclic task graph, by simultaneously minimizing makespan (total execution time), total risk, and the total cost to deploy the network.
We compare and contrast different optimization techniques (leveraging the generality of the previously proposed NSDC (network synthesis for dispersed computing) framework) including brute force and simulated annealing.