Show IH: Publishing the MDVRPTW Benchmark Results of GSL Solver
As part of the ongoing publication of the (Geo Strategy Logistic) GSL Solver benchmark portfolio, today I'm sharing the benchmark results of the MDVRPTW (Multi-Depot Vehicle Routing Problem with Time Windows) module.
Like the previous routing modules, this work follows one core principle:
Deterministic optimization.
No stochastic search.
No per-instance parameter tuning.
The same input always produces the same output.
Benchmark Coverage
The published benchmark results include:
✅ Cordeau MDVRPTW benchmark (20 public instances)
✅ Vidal large-scale benchmark (12 public instances)
32 public benchmark instances
Verification
Every solution was independently verified for:
• Capacity constraints
• Time-window constraints
• Depot assignment
• Complete customer coverage
Large-Scale Validation
The engine was also evaluated on Vidal's large-scale benchmark instances containing up to:
• 960 customers
• Multiple depots
• Strict time windows
Execution times ranged from 1.67 seconds to 19.08 seconds, depending on instance size.
Real-World Commercial Validation
Beyond public benchmarks, the engine was validated using a commercial routing dataset with asymmetric road networks.
Results:
• Original routing distance: 537.43 km
• GSL routing distance: 368.35 km
• 31.4% reduction
• 7 vehicles
• Runtime: 0.0078 seconds
—all executed on an Android phone.
The Bigger Idea
Many routing papers stop at benchmark performance.
My long-term goal is slightly different.
I'm trying to build a single deterministic optimization architecture that can consistently solve multiple routing problem classes while remaining reproducible, verifiable, and deployable on ordinary hardware.
With the publication of benchmark
results across:
CVRP
VRPTW
MDVRP
MDVRPTW
the first public benchmark portfolio of GSL Solver is now publicly documented.
There is still a long way to go, but this is another step toward that vision.
Repository:
https://github.com/CT1-deMo-goG/gsl-mdvrptw-engine
The 31.4% cut on the real commercial dataset is your headline, not the benchmarks. Nobody buying this has heard of Cordeau. Lead with 537km down to 368km on 7 vehicles, that's a number an ops manager can take to a budget meeting.
Thank you for the insight, Producktive. You're absolutely right—the operational impact is what truly resonates with decision-makers. I often get caught up in the technical benchmarks, but presenting the ROI in a language that budget holders understand is crucial. I'll definitely pivot my focus to highlight that 31.4% efficiency gain in future communications. Really appreciate the perspective!