A quiet Saturday: no new papers, and a look at optimization in routing
The day in optimization
No new papers or releases were posted today, so there is no theory to cover. The practice items below are not from this week. They are earlier published deployments, included as background. Treat the figures as the sources' own claims.
Theory
Nothing new today.
In practice
Routing at DHL Supply Chain. An INFORMS paper on large-scale routing at DHL Supply Chain describes an integer-programming system for planning transport routes. The summary reports about $117 million in estimated savings for DHL Supply Chain North America and its customers over roughly 2.5 years starting in 2020. It also reports a 20% increase in win rate and at least 0.1 megatons less CO2. These are the authors' estimates, not independent audits. In plain terms, integer programming picks yes/no decisions, such as which truck serves which stops. At this scale the challenge is keeping those models solvable in practical time.
Open-source GPU solvers. NVIDIA announced that it is open-sourcing cuOpt, its GPU-accelerated engine for decision problems such as routing. The code is also listed in the coin-or/cuopt repository on NVIDIA's developer blog. The announcement date did not appear in the search results, so this is not news from this week. Open code makes it easier for teams to try GPU solving on their own routing problems.
Worth watching
- Vendor case studies, such as Gurobi's Swissport scheduling study, report that staff preferences are met 95–100% of the time and that it saves over $1 million a year. These figures are Gurobi's claims. The study is older, so check any such number against the original source.
Sources: - Innovative Integer Programming Software and Methods for Large-Scale Routing at DHL Supply Chain - NVIDIA Open-Sources cuOpt - Accelerate decision optimization using open source NVIDIA cuOpt - Swissport: automating airport staff scheduling
Try ProvenDecisions free: type, photograph or say an optimization problem; every answer is checked by a solver.