GA and PSO-based resource scheduling for orchestrated, distributed computing

A new distributed computing architecture, Dynamic Virtual Private Network (DVPN), is introduced. The DVM (Dynamic VPN Manager) works as the Autonomous System (AS) administrator in the DVPN system to perform resource scheduling and liaise with the underlying connection management. The approach combines on-demand reservation of both the communications infrastructure and various higherlevel processing facilities. This enables support of orchestrated computing where a complex job can be considered to be a VPN community. This job may be decomposed into tasks to be located at various distributed processing sites. Data can flow between them rather like a production line, in order to deliver the finished "product" to chosen end hosts. Two variants of a resource-scheduling algorithm are proposed for job scheduling in the DVPN system. Genetic Algorithm (GA) and Particle Swarm Optimisation (PSO) mechanisms are considered for use within the optimization process. Simulation results show that both approaches are feasible. The authors then compare the performance of GA against PSO in this dynamic VPN environment to compare their suitability.