How to run CPO benchmark

Requirements

IBM ILOG CP Optimizer (we use the OPL interface)
NodeJS (https://nodejs.org) or bun (https://bun.sh/)

Instructions

transform.mjs

The file transform.mjs transforms the jobshop instances from the json format to a format understandable by OPL (.dat) and puts them in ./data It also generates the file intances.dat that describes the main parameters of the benchmark (time limit, instances to be ran) that serves as input of benchmark.mod

Before launching transform.mjs

create the folder ./data
update the variables time_limit and hardware
update the filters

The filters are based on the content of bks.json

const filter = {
    family : new Set(["dmu", "ta"]),
}

Generates all the instances from families dmu and ta

const filter = {
    status : new Set(["open"])
}

Generates all the instances that are marked as open in bks.json

run transform.mjs with the instruction

node transform.mjs

instances.dat

You can also edit instances.dat if you don’t want to do simple changes without running transform.mjs again

benchmark.mod

The file benchmark.mod controls the loop that

reads and solves an instance
generates the file results.json
generates the file solutions.json

it takes as input the files instances.dat and requires jobshop.mod to be present in the same folder

Run benchmark.mod either by

creating in OPL Studio a “run configuration” (in an existing OPL project) containing benchmark.mod and instances.dat and running it via the UI
running OPL from the command line

jobshop.mod

The file jobshop.mod is the OPL model for the jobshop. It can be tested independently of the benchmark controlling script by creating a “run configuration” (in an existing OPL project) containing jobshop.mod and an instance like ft06.dat

results.json

The benchmark script creates a file results_CPO_{date}.json and for each problem solved an entry of the form

{
    "instance" : "abz5",
    "lb" : {
        "value":1234,
        "date":"2026-6-21",
        "solver":"CPO",
        "hardware":"Intel 11th Gen Core i7-1185G7",
        "time": 1,
        "certificate":"no" 
    },
    "ub" : {
        "value":1234,
        "date":"2026-6-21",
        "solver":"CPO",
        "hardware":"Intel 11th Gen Core i7-1185G7",
        "time": 1,
        "certificate":"no"
    }
}

solutions.json

The benchmark scripts writes the best solution found for each instance in solutions_CPO_{date}.json with the following format

{
    "instance":"abz5",
    "makespan":1234,
    "solver":"CPO",
    "date":"2026-6-21",
    "machine_4" : [7, 4, 5, 0, 1, 6, 3, 8, 2, 9],
    "machine_8" : [2, 4, 0, 9, 5, 7, 6, 1, 8, 3],
    "machine_6" : [7, 1, 8, 5, 2, 0, 4, 3, 9, 6],
    "machine_5" : [1, 5, 7, 2, 0, 4, 3, 8, 6, 9],
    "machine_1" : [5, 6, 2, 7, 9, 3, 8, 0, 4, 1],
    "machine_2" : [3, 7, 1, 4, 5, 8, 0, 6, 2, 9],
    "machine_9" : [2, 4, 9, 7, 5, 0, 3, 8, 6, 1],
    "machine_7" : [3, 6, 8, 7, 9, 5, 4, 2, 0, 1],
    "machine_0" : [8, 2, 9, 4, 7, 6, 3, 1, 0, 5],
    "machine_3" : [4, 1, 7, 9, 8, 6, 3, 5, 2, 0]
}

For the moment solutions are not double-checked by a separate function. In order to validate the solution in a site like optimizizer (https://optimizizer.com) you need to reorder the machines and keep only the list of numbers

2 9 4 7 6 3 1 0 5
6 2 7 9 3 8 0 4 1
7 1 4 5 8 0 6 2 9
1 7 9 8 6 3 5 2 0
4 5 0 1 6 3 8 2 9
5 7 2 0 4 3 8 6 9
1 8 5 2 0 4 3 9 6
6 8 7 9 5 4 2 0 1
4 0 9 5 7 6 1 8 3
4 9 7 5 0 3 8 6 1