High-performance transportation optimization through parallel and collaborative methods
09/2009 — 07/2012. Project ended.
- the Optimization and Heterogeneous Computing groups of SINTEF ICT, Oslo, Norway
- The Dolphin Group, INRIA Lille, France
- CIRRELT, Montreal, Canada
- ITMMA, University of Antwerp, Belgium
In collaboration with research groups abroad, SINTEF has developed highly competitive methods for rich, industrial variants of the Vehicle Rout ing Problem and the Dynamic Shortest Path Problem. The methods have been implemented in vehicle routing and route finding tools, e.g., the Spider technology, and disseminated to industry through vendors. Hence, they improve transportation efficiency in ma ny organizations.
Until recently, the performance of transportation optimization software has increased rapidly through a combination of research based improvement of optimization methods and the general increase of CPU clock frequency. The latter has stag nated due to technological limits. Increasing power is now instead offered through an increasing number of cores. Graphics processors (GPUs) become more programmable and show a drastic increase of performance for data-parallel computing. In the near futur e, commodity processors will be heterogeneous and combine a high number of cores and programmable GPUs. To exploit these developments, it is now absolutely necessary that new and parallelized transportation optimization methods are developed.
In COLLAB, w e develop new parallel and collaborative optimization methods, starting from existing, high performance methods. The methods will fully utilize current multi-core and heterogeneous hardware platforms, and automatically benefit from the future performance increase of such platforms. The COLLAB consortium consists of the Optimization and Heterogeneous Computing groups of SINTEF, plus three renowned research groups in parallel computing, discrete optimization, and transportation from abroad. One postdoc has been employed at SINTEF.