2098. Current progress in highly parallel ensemble solvers instantiated by UG
Invited abstract in session WB-43: Parallelize now!, stream Software for Optimization.
Wednesday, 10:30-12:00Room: Newlyn GR.07
Authors (first author is the speaker)
| 1. | Yuji Shinano
|
| Optimization, Zuse Institue Berlin |
Abstract
The Ubiquity Generator Framework (UG) is a powerful tool to parallelize state-of-the-art solvers that implement the latest algorithms. There are several successful results in solving previously unsolved Mixed Integer programming(MIP) instances, Steiner Tree Problem(STP) instances, Quadratic Assignment Problem(QAP) instances, etc. for the first time by using supercomputers. UG was originally a tool to parallelize branch-and-bound based solvers, but, starting from version 1.0, it was generalized to parallelize any kind of solver that needs to share information among running parallel solvers flexibly. Moreover, the flexibility of UG allows us to parallelize solvers that implement different algorithms as libraries that can run together and cooperate in parallel. We call it a highly parallel ensemble solver. In this talk, we present the latest status of such highly parallel ensemble solver development.
Keywords
- Parallel Algorithms and Implementation
- Software
- Combinatorial Optimization
Status: accepted
Back to the list of papers