wiki:ProjectManagementIdeas

Version 85 (modified by Jörg Viola, 15 years ago) (diff)

Notification of SchedulingToolsPlugin

Trac is strong in basic, individual and small-team task management but lacks features for heavy-duty project management a la Microsoft Project, Project Manager Workbench, etc. This page discusses those missing features and how they can best be realized.

Requirements and Definitions

Project management support software should help us answer a few basic questions:

  • When will my project be done? (That is, what is the forecasted completion date.)
  • How much Earned Value has the project achieved?
  • How do incurred costs compare to Earned value (Cost Performance Factor).
  • Am I on schedule? (That is, is the percent complete for the overall project where it should be at this time.)
  • What tasks are on the critical path? (That is, what tasks have a positive or negative effect on the overall schedule if they are early or late.)
  • How are my resources being utilized?

To answer these questions, the system must know about:

  • Tasks to be done
  • Dependencies between tasks
  • Resources available to do tasks
  • Assignment of resources to tasks
  • Deadlines for task or project completion

Some of this information is part of the Trac core, some is provided by existing plugins, and some must be added to support project management. None of this information is antithetical to Trac's design and implementation.

Properties

A project management system must track properties of (information about) tasks, resources, and milestones.

Task Properties

Answering the questions listed above requires recording certain properties of a task.

Original Work Estimate
The amount of work believed to be needed to complete the task as of before the task was started. This is not, necessarily, the time it will take to complete the task (duration). The task may be unassigned (and will take forever to complete), may have a resource assigned only part-time (so it will take longer to complete than the amount of work indicates), or have multiple resources assigned (so it may take less time to complete than the amount of work indicates). Typically work will be expressed in hours of effort of a resource.
Current Work Estimate
The total time the task is expected to take, as of now. Like Original Work Estimate, but may be modified.
Expended effort
The amount of work expended against this task so far.
Assigned resources
The individuals who will work on this task.
% Effort
Per resource, the amount of the resource's time that will be spent on this task. If the resource is working on a 2-day task 50% of the time, we expect it to take 4 days.
% Complete
The amount of the task that has been done already. If there are 10 subtasks with 10 hours original work estimate and 6 are done, the task is 60% complete.
Dependencies
Other tasks that affect when this task can be started or finished.
Task Hierarchy
A "Work Breakdown Structure" is often used to give a hierarchical organization of all tasks on a project. For example, interior painting of a new house might be broken down into paint the living room, paint the dining room, etc. In a WBS, task 1.1.2 would have subtask 1.1.2.1 and 1.1.2.2. We propose to express the hierarchy but not force explicit numbering. We propose that each task have a single parent, which is the next level up in a WBS.

Dependencies

Task dependencies can be quite complex. Project management generally involves four types of dependencies between tasks or activities:

Finish-to-Start (FS)
Task B cannot start until task A finishes. This is the most common. Sometimes A is referred to as B's Predecessor. When building a house, the walls cannot be put up until the foundation is complete.
Start-to-Start (SS)
Task B cannot start until after task A starts (though they may finish independently). When building a house, interior painting cannot begin until the dry wall is started. (Each room can be painted as soon as its dry wall is up.)
Finish-to-Finish (FF)
Task B cannot finish until task A finishes (though they may begin independently). When building a house, inspections cannot be completed until construction has completed (though parts may be inspected before the whole job is done).
Start-to-Finish (SF)
Task B cannot finish until Task A starts. When building a house, the builder may begin getting interim payments during construction but payment will not be completed until the owner can start moving in.

While useful, the FF and SF dependencies are somewhat more esoteric than the others and implementation may be deferred.

NOTE: The relationships are directional and inverting them is not straightforward. If A has a FS dependency on B, it is not true that B has an SF dependency on A.

Resource Properties

In general a project resource may be many things such as a person working on a task (a carpenter or a software developer) or a source of supplies needed for a task (cement or paint or computer systems).

A resource may have a name and typically has some availability. An employee may have responsibilities in two departments and only be available for certain work 50% of the time or a vendor may only be able to supply a certain amount of cement or paint or CPU cycles.

We don't seek to make Trac a completely general project management solution and will -- at least for now -- limit resources to people and assume they are available full time.

Milestone Properties

The essential property of a milestone is its date. Trac tickets which are part of a milestone are implicitly due by the milestone date. In general project management, milestones have FS dependency with some some tasks. Together with dependencies between tasks, this dictates when tasks must be done to keep a project on schedule.

Task Scheduling

There are two fundamental ways to approach scheduling. One is to assume that all tasks are small and to divide total estimates by resource rate. Then, one compares earned value to cost, computes CPF (Cost Performance Factor), and computes an estimated end time. This is the far simpler approach, but fails to capture tasks that only some people can work on and other complexities. The other approach is to create an actual plan that shows when each resource will work on tasks.

The basic answer to "When will my project be done?" is generally displayed in a Gantt chart which shows tasks, their dependencies, their duration (scaled by resource availability), and milestones. Two threads on the Trac Users mailing list, suggest that a Gantt chart is a fundamental requirement for project management but scheduling tasks is even more fundamental and a necessary precondition for producing a Gantt chart.

Consider a grossly-simplified project to design a new electronic device. This project might have the following tickets:

IDSummaryMilestone
1Design the SchematicNew device
2Design the MotherboardNew device
3Design the Daughter boardNew device
4Design the packagingNew device
5Write firmwareNew device
6Assemble unitsNew device

With this simple task list -- supported by core Trac tickets and milestones -- we'll know when we are done because all the tickets are closed but we have no way to project when that will be. If all the tasks could be done concurrently, the project length is the length of the longest task. And since Trac assigns a date to the milestone, we can sort of work backwards from that to determine when work needs to start for the milestone to be met. However, core Trac has no way to record the time that a task should take. We need a field to hold Work.

IDSummaryWork
1Design the Schematic24h
2Design the Motherboard8h
3Design the Daughter board8h
4Design the packaging16h
5Write firmware40h
6Assemble units8h

From here we can guess that the shortest the project could be is one week. However, we know we cannot begin circuit board layout without a schematic. We need a field to hold Predecessors.

IDSummaryEstimatePredecessors
1Design the Schematic24h
2Design the Motherboard8h1
3Design the Daughter board8h1
4Design the packaging16h
5Write firmware40h1
6Assemble units8h2,3,4,5

For such a short task list we can manually inspect and see that assembly (8h) follows firmware (40h) follows schematic (24h) and the shortest project time is 72 hours (9 man-days) and we can take nine days from the milestone date to know when we have to start work to finish on time. However, a project of any complexity may have many, many more tasks and manual inspection is impractical.

We can begin to do more effective project management if we have a task scheduler than can work backwards from the milestone, consider work and dependencies, and calculate task start and end dates and project start date.

If the milestone for this work is May 11, the schedule might look like the following.

IDSummaryWorkPredecessorsDueStart
1Design the Schematic24hMay 5, 2009May 3, 2009
2Design the Motherboard8h1May 10, 2009May 10, 2009
3Design the Daughter board8h1May 10, 2009May 10, 2009
4Design the packaging16hMay 10, 2009May 9, 2009
5Write firmware40h1May 10, 2009May 6, 2009
6Assemble units8h2,3,4,5May 11, 2009May 11, 2009

This is computed as follows:

  • Assemble units is the last step and is due when the milestone is scheduled
  • Firmware, packaging design, and the circuit board all must be done before assembly or 1 day (the length of the assembly task) before the milestone
  • Schematic is due the length of the circuit board task before the circuit board's end

To finish on time, the project must start on May 3, the earliest start time for any task.

However, the preceding schedule assumes that the motherboard and daughter board are independent tasks. If there is only one board designer, only one board at a time can be worked on. If we assign the same resource to the board designs, either one must be done before the other can begin (even in the absence of an FS dependency) or the time for both must be extended and the resource spread between them concurrently.

With 100% applied to each task, the motherboard and daughter board are worked on serially.

IDSummaryWorkPredecessorsResourceDueStart
1Design the Schematic24hMay 5, 2009May 3, 2009
2Design the Motherboard8h1MontyMay 9, 2009May 9, 2009
3Design the Daughter board8h1MontyMay 10, 2009May 10, 2009
4Design the packaging16hMay 10, 2009May 9, 2009
5Write firmware40h1May 10, 2009May 6, 2009
6Assemble units8h2,3,4,5May 11, 2009May 11, 2009

With 50% applied to each task, the board designs are worked on in parallel but take twice as long.

IDSummaryWorkPredecessorsResourceDueStart
1Design the Schematic24hMay 5, 2009May 3, 2009
2Design the Motherboard8h1Monty (50%)May 9, 2009May 10, 2009
3Design the Daughter board8h1Monty (50%)May 9, 2009May 10, 2009
4Design the packaging16hMay 10, 2009May 9, 2009
5Write firmware40h1May 10, 2009May 6, 2009
6Assemble units8h2,3,4,5May 11, 2009May 11, 2009

Calendar

The schedule above isn't accurate unless we have a 7-day work week. If we assume that the work week is Monday to Friday with Saturday and Sunday off, firmware has to finish Friday, May 8 and schematic has to finish Friday, May 1 so the project must begin April 29. Applying calendar information, our task list looks like the following.

IDSummaryWorkPredecessorsResourceDueStart
1Design the Schematic24hMay 1, 2009April 29, 2009
2Design the Motherboard8h1MontyMay 7, 2009May 7, 2009
3Design the Daughter board8h1MontyMay 8, 2009May 8, 2009
4Design the packaging16hMay 8, 2009May 7, 2009
5Write firmware40h1May 8, 2009May 4, 2009
6Assemble units8h2,3,4,5May 11, 2009May 11, 2009

Gantt Charts

Dependency and Duration

Once we have a task scheduler, we can simplify presentation of task scheduling with a very basic Gantt that showed just tasks and their dependencies and the resultant schedule.

A simple Gantt chart

Disregarding any resource constraints, this chart shows

  • circuit board design and firmware development follow schematic design
  • package design can proceed independently
  • assembly can't be done until the circuit boards and packaging are ready
  • the product can't ship until it's assembled and has firmware.
  • the overall project will take 9 calendar days

This requires only that the ticket have some information about estimated effort or duration of the task and simple FS task dependency.

Progress

If we also track work on tasks, we can monitor project progress at a glance.

Gantt chart with progress bars

Resource Allocation

When a resource is not available 100% of the time, tasks take longer than the work suggests. By assigning effort to resources, we can chart this time dialation.

Duration longer than work due to resources

When the hardware and software engineers are only available half time, the project gets delayed.

Critical Path

Some tasks affect project schedule and some do not. The tasks which affect the schedule are on the critical path.

Critical path

Packaging design and circuit board layout are so short -- relative to other activities -- that speeding them up doesn't really help the project. But if we could shorten schematic design or firmware, every day we take out of one of those activities we also take out of the project schedule.

Sub-tasks

To manage the complexity of the chart, we can break tasks down into sub-tasks and show various levels of detail.

Rolling up subtasks - 1

Rolling up subtasks - 2

Progress Line

A Gantt chart often includes a "today" line to show where a project stands.

Progress line

Bumps to the left indicate tasks that are behind schedule.

Slack Time

A Gantt chart may also show slack time for tasks, how much they can slip before affecting the overall schedule.

Slack time illustration

The black bars extending from the bottom of circuit board and packaging design indicate how much those tasks can slip without affecting deliverable.

Should talk about lag here

Resource Utilization

There are several views of resource utilization which may be helpful:

  • utilization over time (80% loaded this week, 110% next week, 70% the week after, etc.)
  • allocation at a time (10% on task 1, 20% on task 2, 50% on task 3, etc.)

Both views may be combined in a sequence of stacked bar graphs. At T1, a resource is 80% utilized with 20% each on four tasks, at T2, one task is complete and the resource is 60% utilized with 20% on each of the remaining three tasks, etc.

Resource utilization chart

Earned Value

To compute earned value, we need unchanged original estimates, and probably also some way to estimate remaining time.

Resource Description and Allocation

We need a proposal for how to describe available resources within Trac. If tasks are tickets and tickets have owners who work on them, it seems reasonable that a resource is a Trac user. There may be information we need to track for a resource that it not part of the user configuration. It might also be necessary or desirable to allow tickets to have multiple owners. Or, we might want to have a way to have a list of (resource,hours) pairs independent of the owner.

Related Work

Trac Plugins

Trac has three Gantt chart plugins:

FlashGantt
seems complex and stale.
GanttChart
is a very simple Gantt chart implemented in a formatted text region. It is manually edited, not data-driven.
TracGantt
has potential but has some funky fields (e.g., Due to Assign) that don't seem to be appropriate to what we want to do with a Gantt chart.

Pale Purple's Virtual Planner is an alternative visualization tool.

MasterTicketsPlugin supports FS dependency (but calls it blocks and blocked by).

The SubTickets page talks about adding composition type dependency. Can this be done in a plugin like MasterTickets or must it be in the core?

TimingAndEstimationPlugin and TracHoursPlugin add estimated and actual hours fields to tickets. There is some disagreement about whether the estimate is the remaining time (so total estimated time is actual + estimate) or total time (so remaining time is estimate - actual). A Gantt chart typically wants estimate and percent complete. It is possible to extract this from either recording convention and we should likely allow either convention via a configurable interface.

TeamCalendarPlugin keeps track of user availability.

May I draw your attention to SchedulingToolsPlugin? I was not aware of this page, but implemented some of its ideas in some kind of prototype. It currently has a scheduler, resource availability and gantt chart in a simple fashion. I would like to enhance it, maybe we can join efforts? -- viola

A fact of life is that Microsoft products like Exchange and Outlook are present in many environments. It would be nice if we could get availability information from Exchange to feed a project calendar. Just reading from the group calendar would probably be enough. That way users would only have to put in their vacations, etc. in one place.

Other work

There is a Google Gadget Gantt chart which might be leveraged for the visualizations once the data is made available (e.g., using TracGViz plugin). That widget is developed and maintained by a company (i.e., ViewPath). Besides users need access to the internet and their projects environments have to be available too so that the iGoogle container be able to retrieve the data to display in the chart. All this means that local deployments (intranets, ...) are not possible.

Planner is an open source project management program.

GNOME Office once included MrProject (which may be another version of Planner). The news of MrProject's death may be exaggerated; GNOME Live! lists a 0.14.4 release in April 2009.

dotProject is an open-source, web-based project management system. A quick review suggests its task management isn't as good as Trac. It has a Gantt chart but wikipedia notes "as of version 2.0 the task dependencies feature is not complete". It has a troubled history but was forked last year as Web2Project.

I reviewed 15 Useful Project Management Tools. It mentions Trac. None of the tools obviously have scheduling in them. Jira sounded promising but doesn't seem to have scheduling. Their chart examples don't include a Gantt. Basecamp is popular and promising:

Basecamp works

For years project management software was about charts, graphs, and stats. And you know what? It didn’t work. Pictures and numbers don’t get projects done. Basecamp tackles project management from an entirely different angle: A focus on communication and collaboration. Basecamp brings people together.

But it is also commercial (which doesn't really help the Trac community). The description above makes it sound like we shouldn't expect a Gantt chart in Basecamp. ;-) I don't see automatic scheduling. They have a nice feature where you can subscribe to milestone updates via iCalendar.

Also:

  • Open Workbench is a Microsoft Windows-based application not suitable for integration with a web-based tool like Trac.
  • OpenProj also appears to be a desktop application. However, their views page may guide development of PM output for Trac.
  • GanttProject's website isn't very informative but it appears to be a desktop application.

Microsoft Project Task Properties

Microsoft Project goes far beyond what is likely to be put into a Trac-based project management solution but it may be helfpul to consider what task properties MSProject records and what their analogs may be in Trac tickets.

The Microsoft Project Task Information property sheet includes:

  • General
    • Name
    • Duration
    • Estimated (Boolean)
    • Percent complete
    • Priority
    • Start date
    • Finish date
  • Predecessors - A list of related tasks and their relationships
    • ID
    • Task Name
    • Type (FS, SS, etc.)
    • Lag (delay between tasks, overlap between tasks or lead if negative)
  • Resources - A list of resources
    • Resource Name
    • Units (% Effort for workers)
  • Advanced
    • Constraint
      • Deadline (a date or N/A)
      • Constraint (ASAP, ALAP, Start no earlier than, etc.)
      • Constraint date
    • Task Type (Fixed Units, Fixed Work, Fixed Duration)
    • Calendar (Standard, 24hr, Night Shift)
    • WBS code (outline number from Work Breakdown Structure such as 1.3 for the third subtask of task 1)
    • Earned value method (% Complete or Physical % Complete)
  • Notes (Like ticket comments, only worse)
  • Custom Fields

However, this doesn't seem to be all the data Project records for a task. If you add a column to the WBS in the Gantt chart, there are many more options. The most interesting is Work which is the amount of work required to complete a task. Only when a resource is assigned 100% are Work and Duration the same. We'd like to be specify the work required for a task and let the system calculate its duration based on available resources.

Roadmap

There seems to be a consensus that grandiose project management features for Trac should be implemented with a combination of plugins which provide useful functionality on their own. The following plan assumes that approach.

  • Tasks will be represented as tickets. Additional, non-core, data will be needed. (In object-oriented terms, you might say that Task is a subclass of Ticket.)
  • We need a way to express WBS relationships, assuming MasterTickets will be used to express dependencies. If we need more than FS dependencies, we will need to extend MasterTickets to express dependency type.
  • TimingAndEstimation can be the basis of recording estimates. The plugin should be extended to support original and revised estimates, and to store default estimates in the database (rather than have rules to use the default value for tickets with tiny estimates).
  • We need a new resource plugin to allow describing resources and their availability. This should support notions of normal work hours, holidays, etc.
  • Basic scheduling can be provided based on core Trac tickets, MasterTickets, and TimingAndEstimation. (This discounts resource conflicts and availability.) A new Scheduling plugin can build on these plugins (or an interface that hides them), add Start and Finish dates to tickets, and schedule going back from a milestone.
  • Once a scheduler exists, a simple Gantt chart can be implemented to show dependencies and schedule. (This might even be a revision or refinement of an existing Gantt.)
  • Additional dependency types can be added, perhaps by enhancing or forking MasterTickets.
  • Additional Gantt display options (e.g., critical path) can be added independent of additional dependency types.
  • The scheduler can be enhanced to take into account resource conflicts.

Design

(I'm a little out of my depth here, designing for Trac internals I don't understand. Hopefully, more experienced Trac developers will guide me. -- Chris)

To add better project management to Trac in a modular fashion requires defining the interfaces that project management implementations rely on. We'll name these with a common prefix "IProject". The 'I' is for "Interface", a common convention in Trac development. The "Project" might be "PM" or "ProjMgr" or something. This is certainly open to discussion.

IProjectTask

Tasks for project management will be based on tickets but an abstract interface allows us to decouple a scheduler or other PM tool from the implementation of non-core ticket features like recording estimates and progress. One user may choose to implement IProjectTask on top of TimeingAndEstimation and another on top of TracHours.

An IProjectTask has the following properties:

id
Numeric ID (Inherent in Trac)
work
Man-hours of work to complete task
risk
Relative risk. An integer from 0 to 100. How likely it is that work is accurate. Zero means no risk; work is certain to be accurate.
priority
Relative priority. An integer.
percentComplete
How much of work is done?
resource
Name of resource assigned to this task
percentEffort
How much of resource's time is spent on the task.
duration
How long will resource take to complete work with percentEffort. For example, a 16-hour task with 50% effort will take 4 days.
assignedStart
An explicit constraint on when the task must start (a date)
assignedFinish
An explicit constraint on when the task must finish (a date)
computedStart
The result of scheduling this task based on constraints. Equals assignedStart if that field is set.
computedFinish
The result of scheduling this task based on constraints. Equals assignedFinish if that field is set.
dependencies
A list of other tasks (by id) and the dependencies of this task on them.

assignedStart and assignedFinish are likely mutually exclusive (that is, only one can be set).

Each dependency specifies:

task
The ID of the task this task depends on (I'd really like a better name here. parent is wrong. origin? other?)
type
Dependency type (FS, SS, SF, or FF)
lag
Offset of this task's anchor relative to the anchor of task. (Lag may be negative.) Whether the anchor for the dependency is the start or end of the task depends on the dependency type. For example, if Task B has a FS dependency on Task A with a lag of 1 day, then Task B starts 1 day after Task A finishes. Or, if Task B has an SS dependency in Task A with a lag of 1 day, then Task B starts 1 day after Task A starts. (lag is not scaled by percentEffort.)

IProjectCalendar

The essential feature of a project calendar is that it knows when work is not done (weekends and holidays) so that task duration can account for that down time.

A slightly more sophisticated implementation would track individual resource availability so that the schedule can account for vacations and such.

An IProjectCalendar should provide the following methods:

startFromFinish(task,finish)
return the date on which task must begin to complete by finish. If task's work is more than a day, this accounts for weekends, etc. so finish - start may be more than task.work.
startFromFinish(task,finish,resource)
return the date on which task must begin for resource to complete it by finish. If task's work is more than a day, this accounts for weekends, etc. so finish - start may be more than task.work. May account for resource's vacation, etc. in addition to the general work calendar.
finishFromStart(task,start)
return the date on which task will complete if started on start. If task's work is more than a day, this accounts for weekends, etc. so finish - start may be more than task.work.
finishFromStart(task,start,resource)
return the date on which task will complete if resource started it on start. If task's work is more than a day, this accounts for weekends, etc. so finish - start may be more than task.work. May account for resource's vacation, etc. in addition to the general work calendar.

IProjectResource

Describes resource availability. May need calendar options for individuals. Some may want progress rates and costs, and some may not - this is surely controversial.

IProjectSchedule

Computing a schedule involves determining the computedStart and computedFinish for a set of tasks taking into account the dependencies between tasks and the resources assigned to those tasks. Where the tasks do not have assignedStart or assignedFinish, the computed schedule prioritizes tasks to keep resources at or below 100% utilization. However, assigned dates may force overloaded resources. These overloads can then be reported and either manually resolved or resolved with a resource leveling module apart from the scheduler.

The following rules control the scheduling:

  • Choosing a method
    • If using As Soon As Possible, tasks without precedessors should be done done today (or the start of the project, if that's later), and their successors when they are done, etc.
    • If using As Late As Possible, tasks without successors should be done at the end of the project and their predecessors done right before they must start, etc.
    • It may be desirable to override the method for individual tasks so exploratory work is done ASAP and clean up work is done ALAP, regardless of the overall method being used.
  • Handle assigned dates
    • It is an error for a task to have an assignedStart and an assignedFinish.
    • If a task has an assignedStart, the computedStart is the assignedStart and the computedFinish is the computedStart plus the duration.
    • If a task has an assignedFinish, the computedFinish is the assignedFinish and the computedStared is the computedFinish minus the duration.

NOTE: Once a task is begun, its assignedStart is set from the computedStart so recomputing a schedule doesn't change its start. We may have to deal with split tasks, however, if a task is begun then put down and resumed later.

  • Handle dependencies (This must be done iteratively for all of a task's dependencies. Several FS dependencies may produce different computedStart dates with the latest one being use. Interaction between other types of dependencies is complex.)
    • If a task has an FS dependency on another task, its computedStart is the other task's computedFinish plus lag. The task's computedFinish is computedStart plus duration.
    • If a task has an SS dependency on another task, its computedStart is the other task's computedStart plus lag. The task's computedFinish is computedStart plus duration.
    • If a task has an FF dependency on another task, its computedFinish is the other task's computedFinish plus lag. The task's computedStart is the computedFinish minus duration.
    • If a task has an SF dependency on another task, its computedFinish is the other task's computedStart plus lag. The task's computedStart is the computedFinish minus duration.

NOTE: A scheduling algorithm which handled only FS dependencies would be a very useful first step.

  • Handle resource limitations
    Tasks assigned to the same resource but with no other dependency between them and no assigned dates must be sequenced to keep from overloading the resource. A comparison function can be used to determine which task should go first (as in many sorting algorithms). sequenceTasks(taskA,taskB) would return -1 if taskA should go first, 1 if taskB should go first or 0 if it doesn't matter. The scheduler should not be aware of the policy implemented in the comparison function. Possible criteria for sequencing the tasks include:
    • priority - More important/urgent work goes first
    • risk - Riskier work goes first
    • work - A function may favor short tasks ("low hanging fruit") or long ones (which inherently have more risk).
    • task type - fix bugs before doing new features
    • percentComplete - Finishing something that's partially done is better than starting something new
    • "fit" - It's better to start a 2-day task on Wednesday and hold a 4 day task for the next week than to break up the longer task across a weekend.

Related research

Scheduling activities in a project is an area of active research in operational research (or operations research, OR) (cf. Project Scheduling) and has been established to be NP hard.

There are several variations on the problem.

RCPSP
Resource-Constrained Project Scheduling Problem
RCMPSP
Resource-Constrained Multi-Project Scheduling Problem
m_PRCPSP
Preemptable RCPSP. A RCPSP where each task may be broken (preempted) m times during scheduling. (Generally, m is limited to 1 both for simplicity of the algorithm and because there is real cost in practical task switching.)

A range of techniques have been brought to bear on RCPSP. They can be broadly categorized as:

  • Exact solutions. Attempts to find an optimal schedule. Due to the polynomial nature of the problem these are only possible or practical for a small number of activities (e.g., a few dozen).
  • Heuristic approximations. Attempts to find a good solution in reasonable time for a realistic number of tasks (e.g., hundreds).
  • Metaheuristics. More abstract approaches such as genetic algorithms, tabu search, simulated annealing, ant colony optimization, and particle swarm optimization.

All of these algorithmic approaches also have other dimensions such as the number of threads that are used, the method they use to prioritize activities, etc. To be practical for implementation as a Trac plugin, it seems likely our implementation should not require heavy-weight, opaque abstractions or multiple threads. Furthermore, we desire an algorithm which can update a schedule as part of a ticket change listener rather than having to completely recompute a schedule for each, individual change.

PSPLIB provides a standard set of program scheduling problems to test the various algorithms performance against one another and the optimal solution.

Kolisch and Hartmann1 tested dozens of algorithms and variations using the PSPLIB data.


Resource allocation and planning for program management Kabeh Vaziri, Linda K. Nozick, Mark A. Turnquist December 2005

Much of the project scheduling literature treats task durations as deterministic. In reality, however, task durations are subject to considerable uncertainty and that uncertainty can be influenced by the resources assigned. The purpose of this paper ...

Stochastic rollout and justification to solve the resource-constrained project scheduling problem Ningxiong Xu, Linda Nozick, Orr Bernstein, Dean Jones December 2007

The key question addressed by the resource-constrained project scheduling problem (RCPSP) is to determine the start times for each activity such that precedence and resource constraints are satisfied while achieving some objective. Priority rule-based ...

Activity scheduling in the dynamic multi-project setting: choosing heuristics through deterministic simulation Robert C. Ash December 1999

And others at ACM.org.

Schedule Scenarios

It might be helpful or interesting to consider saving schedules or scenarios. If we stored resource assignment, start and finish data in a table keyed by schedule and ticket, we could store multiple possible schedules and choose to display different ones. An initial release could have only a single schedule and no facility for creating alternatives. A later refinement could add scenario support. If a start or finish date was configured for a task, the schedule would copy that data and not allow editing or recomputation of those dates; other tasks would flow around that fixed time. If a task lacked a start or finish date, the scheduling of that task would be fluid and computed by the scheduler.

Disclaimer

The bulk of this page was prepared by Chris Nelson who makes no claim to be certified project planner or anything of the sort. It is based on his understanding of project management after 25 years of software development. Other developers have also offered some hints and useful comments (according to their experience) in order to enhance this page, and get a better understanding of the features and requirements needed.

Contributors

  • Olemis Lang
  • Greg Troxel


  1. 1. Kolisch, R. & Hartmann S. "Experimental investigation of heuristics for resource-constrained project scheduling: An update" European Journal of Operational Research 174 (2006): 23-27.


Old design notes

(I'm reworking these into a better flow for the whole document.)

Component Data

Each Gantt chart component or feature requires data to support it. For Trac-based project management, this data is often in addition to the core ticket data.

In general, we might expect to have multiple resources assigned to a task but if we use a Trac ticket as the basis for a task, we have only a single resource, the ticket owner.

Core Fields

Core ticket fields include relevant to project management include:

  • owner (who is working on it)
  • milestone (what milestone this ticket is part of)
  • status (new, assigned, ..., closed)
  • summary (short description)

Core milestone fields relevant to project management include:

  • name
  • due (date and time)
  • completed (treated as a Boolean: open or closed)

Additional Data

For project management, we must have the following additional information about a task/ticket:

  • estimate (expected work required to complete task)
  • percent complete
  • dependencies
  • percent effort

NOTE: While Gantt charts typically show percent complete, this can be computed from time remaining vs. total estimate or time worked vs. total estimate. We will not, necessarily, store percent complete. It is intuitive for display and analysis but difficult for data entry.

A Gantt plugin for Trac should analyze ticket dependencies and produce an interactive, explorable Gantt chart showing task progress and project status.

By "project", I mean a set of related milestones. Perhaps a software project has a Design Phase, an Alpha Release, a Beta Release, and a General Release. Each phase would have a milestone with a target date and tickets to complete the work for that phase. There must be a way to specify which milestones to include either listing explicitly or by giving a pattern or substring filter.

What do we need in the back end ?

The chart should be able to show all tickets (though that may be a very complex chart) or alternately all those tickets matched by a (query | report) or any other ticket group provider.

To effectively display project progress, tickets must have estimated and actual times as in the TimingAndEstimation plugin.

It is also desirable to have loop detection to error-proof the tool used to create dependencies.

It would be nice to be able to choose an As Late As Possible or As Soon As Possible algorithm for laying out tasks.

What do we need in the user interface ?

A "chart these" button on a report page would be very nice.

Tasks in the chart should be links to the tickets they represent (preferentially opened in a separate tab or window).

Milestones should be displayed in the chart as well.

The chart should support interactively "folding" groups of tasks to hide detail.

Zooming in order to see the WBS in more detail.

Panning or scrolling in order to display the schedule at a different time.

Each task's bar in the chart should show % complete.

The chart should have a Today line.

An accompanying report or tool should aid in resource leveling by (at least) showing overcommitted resources.

Visual indications about delays, task completion and other significant events.

It would be nice to be able to create task dependencies graphically using the Gantt chart as a GUI or to change a milestone date by dragging it along the chart.

Design guidelines

In order to agree about how to implement this functionality, the very first step should be to describe the possible architectures (i.e., how the different components are going to interact with each other since the moment the request is sent by the client until the Gantt chart is rendered by the browser) that will be considered.

Components

In order to add flexibility to the final solution, the MVC pattern should be considered during its development. Hence the following components need to be implemented in order to add Gantt charts to Trac.

The Model

It provides the data in the Gantt chart and is not just a single element but a set of components like the following :

  • The ticket models, which provide data about the issues found in the project environment
  • Trac query mechanism which provides a way to select (filter) the tickets being shown.
  • Other plugins used to track ticket dependencies
  • Other plugins used to track task/subtask relationships
  • Other plugins used to estimate the time needed to finish the given tasks
  • Maybe one (a few) component(s) whose role is to provide the functionality specific to Gantt charts, thus acting as a central access point and a mediator so as to put all the other components together.

The View

It should be a JavaScript component (is this possible ?) responsible for rendering the data provided by the model. Besides it might support other dynamic features

The Controller

¿What do we need this for?

Implications

The aforementioned separation should allow the replacement of any of the components involved. Hence the following modifications should be made easily (the list being intentionally incomplete) :

  • Replace the JS-based view by another made with Flash.
  • Use a custom, or specific estimation strategy.

Attachments (10)

Download all attachments as: .zip