Johnson's rule sequences n jobs on two machines so the last job finishes as soon as possible. It's shop-floor math, not a job-board feature. And on a site about job search apps, that distinction matters more than usual.
The rule works inside one narrow cell: every job runs Machine 1 then Machine 2, the times are known ahead, nobody cuts the line, and makespan is the only score that matters. That cell is narrower than most U.S. job shops.
To be honest, the phrase sounds like recruiting. So why does this URL exist on a job-search site? Production schedulers still get asked about the method in interviews, and it's easy to stretch the claim past what the math supports. That's the gap this page fills.
What Johnson's rule actually does
S.M. Johnson published the method in 1954. You have n jobs. Each one needs a stretch of time on the first machine and then another stretch on the second.
Idle time on Machine 2 is the enemy. That's the whole fight. Short first-machine jobs go early so the second machine gets fed fast, and short second-machine jobs go late so you're not stuck with a long finish after Machine 1 has already gone quiet.
You sort. You never enumerate every possible sequence. A worked two-machine explanation frames it as a partition plus a sort, which is a fair way to picture it.
Use this procedure once the times are known:
- List every job with its Machine 1 time and its Machine 2 time.
- Split the jobs. A job lands in set A if its first-machine time is less than or equal to its second-machine time. It lands in set B if the second machine is strictly shorter.
- Sort set A by increasing first-machine time. Sort set B by decreasing second-machine time.
- Run set A first, then set B.
- Read the makespan off the last finish on Machine 2.
There's a second version of the same trick. Find the smallest time left anywhere on either machine, then park that job at the front if the small number sits on Machine 1, or at the back if it sits on Machine 2. Repeat until the jobs run out. You land on the same sequence either way. Notes on scheduling two machines show why an adjacent pair that's out of order is never better.
A small example you can work by hand
Here are five jobs on a cut-then-finish pair.
| Job | Machine 1 | Machine 2 |
|---|---|---|
| A | 5 | 2 |
| B | 1 | 6 |
| C | 9 | 7 |
| D | 3 | 8 |
| E | 10 | 4 |
Set A holds B and D, since their first-machine times are shorter or tied, ordered B then D. Set B holds C, E, and A, sorted by falling finish time: C, E, A. Final order: B, D, C, E, A.
Five jobs means 120 permutations, and you're testing exactly none of them. Thing is, that Gantt chart only stays honest if the route never flips and nobody drops a hot job into the run. A two-stage sequencing example cut a teaching makespan from 101 minutes in arrival order to 95 minutes with a Johnson sequence. That gap is for that dataset.
When the rule is actually optimal
The guarantee covers a two-machine permutation flow shop, full stop. A job shop where each job has its own route is a different problem.
Every one of these has to hold:
- Exactly two machines, and every job uses both in the same order
- All jobs ready at time zero
- No preemption. A started operation runs to the end.
- One job on a machine at a time
- Deterministic processing times
- Makespan as the objective, not tardiness or average flow time
People still write "two-machine job shop" when what they mean is every job runs M1 then M2. Self-learning material on that two-machine case uses exactly that wording.
If your shop matches that shape and makespan is the metric, start with Johnson's rule. Don't open with FCFS. Flow-shop makespan notes give a lower bound worth knowing: the larger of total Machine 1 time plus the smallest Machine 2 time, and total Machine 2 time plus the smallest Machine 1 time. Hit that bound and you can stop.
Where it stops helping
Three or more machines and the optimality proof is gone. People glue stages together and run Johnson's rule as a heuristic when a middle machine is dominated, which is fine as long as you call it a heuristic. Flexible routes, breakdowns, and due dates need another method.
How this shows up in U.S. hiring
Turns out most scheduler postings never name Johnson's rule. They name sequencing, capacity, and the plant's ERP. That's the real checklist.
Reciting 1954 won't carry a posting by itself. A skills line that only says "Johnson's rule" reads thinner than one that names the two machines, the fixed route, and makespan as the objective you actually ran. Say what you scheduled.
Employers should ask for a five-job sequence, and for one case where the candidate would refuse the rule. That's a better filter than a keyword hit.
Keep the method in your interview kit for operations roles, and leave it off a software or recruiting resume. Echo whatever scheduling tools the posting lists. Johnson's rule is supporting knowledge, and it should behave like it.
A working check before you claim it
Run this before you teach the rule or add it to LinkedIn.
- Are there two machines, same route for every job?
- Are times known, and is makespan the metric?
- If you have more machines, are you calling it a heuristic?
- Can you draw a five-job Gantt in five minutes?
Any no means you describe the dispatcher or ERP rule you really used instead. Hiring teams prefer that.
Take one past role and write the routing plus the objective in a single sentence. Then decide whether Johnson's rule belongs in that sentence. If it does, add the sequence logic. If it doesn't, prep a shop-floor story that matches the plant you actually worked.