Job sequencing rules: which job goes first, and why it matters

The total work on a machine does not change with the sequence. Five jobs that need 23 hours still need 23 hours. What changes is when each job finishes, and so which jobs make their due dates. On a bottleneck machine, that is the difference between a good week and a week of apologetic phone calls.

All five jobs are waiting at the start of the day, and the machine runs them one after another. Times are in hours from now.

Run the quickest job first. Sequence: B, D, E, A, C.

Result: two late jobs, 12 hours of total lateness, and an average finish time of 10.8 hours, the lowest of the three rules. SPT clears the queue fastest and keeps work in progress low. Its weakness is that long jobs keep getting pushed back, and here job A, due soonest after B, ends up the most late.

Run the job due soonest first. Sequence: B, A, E, D, C.

Result: two late jobs, but only 7 hours of total lateness, and no job more than 5 hours late. EDD is the rule that best protects due dates on a single machine, which is why it is the sensible default when on time delivery is what your customers judge you on.

Suppose job C is for your most important customer, and you run it first, then the rest by due date. Sequence: C, B, A, E, D.

Result: four late jobs and 30 hours of total lateness, so that a job with ten hours of slack could finish early. Priority rules are real and sometimes necessary, but they are expensive, and it is worth seeing the price before paying it. Under EDD, C finished at 23 hours, five hours after its due date, so putting it first did protect it, but the other four jobs paid for that with 28 more hours of lateness than under EDD, and total lateness rose from 7 to 30 hours.

• Earliest due date: the default when on time delivery matters most, which is most make to order plants.
• Shortest processing time: when the queue itself is the problem, work in progress is piling up, and due dates have slack.
• Customer priority: for genuine rush orders, applied to as few jobs as possible, and with the effect on the others checked first.

In practice, plants mix them: EDD for the bulk, priority for a handful of flagged orders, and grouping of similar jobs on machines where setups are long.

With one machine, the rule decides everything. With a real routing through several work centers, a job's finish time also depends on when the previous operation finished and on the queue at every later machine. A rule that is best on one machine is not automatically best for the plant. The practical answer is to schedule the whole plant with finite capacity under each rule and compare the late order lists, rather than reason it out by hand. See machine scheduling software for the queue on each machine, and advanced planning and scheduling software for the engine that places every operation.