You have a project or program that will take a long time to finish and you only have a fixed maximum time slot to work on it every day. The project is structured and divided into a series of smaller tasks or milestones such that:
- The tasks are ordered and performed sequentially. This means a following task cannot start until the previous task is finished.
- The duration of each task differs. Some tasks take longer than others but each task lasts shorter than the daily time slot allowed. Some tasks may be so short that it's possible to finish several within the same time slot.
- Tasks have overhead time penalty. The first and last task completed in a time slot incur set-up and tear-down time penalty, respectively. The penalty amount depends on the task.
|Task||Execution Time||Set-up Time||Tear-down Time|
All time units are in minutes. Set-up and tear-down time are the time penalty for context switching tasks in and out of the time slot. In real-life, these could be ceremonious activities such as booting up the machines, push or pulling source code from version control, reading or documenting code, loading or persisting large data set to or from memory, powwows, soaking up chakra, etc.
There are many ways of scheduling the tasks. Some are valid and some are not. For example, if you choose task A, B and C on the first day, the set-up time is 10 (A's set-up time). Tear-down time is 5 (C's tear-down time). Task A, B and C take 50, 30 and 40 minutes to execute, respectively. The total time to complete A, B and C is 10 + 50 + 30 + 40 + 5 = 135 minutes. Since your maximum daily time slot is 120 minutes, this combination exceeds the time slot so it's not a valid choice.
As an example of a valid scheduling, you can assign A and B to the first day, C and D to the second, and E to the third. The total time to complete the project is:
(A + B) + (C + D) + E = (10 + 50 + 30 + 0) + (15 + 40 + 20 + 10) + (5 + 30 + 5) = 215 minutes
As much as we programmers like to program, we like downtime, too. The quicker we finish the project the better. So how do we find the best scheduling assignment? There is a simple and clever solution to this thanks to Prin et al  from the seemingly unrelated problem domain of vehicle routing. I'll explain the intuition behind the idea. For those who are looking for an efficient labeling algorithm implementing the idea, please refer to Prin et al .
First, we'll represent the scheduling problem with a graph: