On-line power savings in a distributed multi-stage router architecture

We focus on a distributed multi-stage software router (MSSR) architecture internally composed by several personal computers (PCs) to overcome scalability and performance issues of software routers (SRs) based on a single PC. Sizing the internal architecture to sustain the peak load may lead to power inefficiency at low loads. This paper presents a power saving scheme to improve the power efficiency of the MSSR by dynamically adapting the size of its internal architecture to the offered load to reduce power needs. The off-line problem is defined as a mixed integer linear programming optimization model, shown to be NP-hard. We propose a differential on-line heuristic to solve the optimization problem when the traffic load changes. The heuristic avoids the complete MSSR reconfiguration of the optimal off-line solution that may lead to forwarding delay increase or service interruption. The performance evaluation shows that the proposed on-line algorithm, that gracefully modifies the internal MSSR configuration, preserves the load proportional power demand characteristics of the optimal offline solution.