[[PageOutline]] 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 tasks. Together with dependencies between tasks, this dictates when tasks must be done to keep a project on schedule. Trac milestones are not first-class objects (they lack history and comments that tickets have). An alternative to Trac milestones is to create a custom ticket type that can be used to set deadlines and have tickets which are required for that ticket scheduled as if they had a milestone with a due date set. To avoid conceptual conflicts with Trac milestones, I've called that ticket type an ''inchpebble''. Trac milestones may still be used to group tickets but a milestone may have many inchpebbles which set intermediate dates. == 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 [http://en.wikipedia.org/wiki/Gantt_chart Gantt chart] which shows tasks, their dependencies, their duration (scaled by resource availability), and milestones. [http://groups.google.com/group/trac-users/browse_thread/thread/83c0b6a248040542?hl=en Two] [http://groups.google.com/group/trac-users/browse_thread/thread/3084796acbc7233c/3f393a18f99cfebd?hl=en&tvc=2#3f393a18f99cfebd}}} 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: ||'''ID'''||'''Summary'''||'''Milestone'''|| ||1||Design the Schematic||New device|| ||2||Design the Motherboard||New device|| ||3||Design the Daughter board||New device|| ||4||Design the packaging||New device|| ||5||Write firmware||New device|| ||6||Assemble units||New 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''. ||'''ID'''||'''Summary'''||'''Work'''|| ||1||Design the Schematic||24h|| ||2||Design the Motherboard||8h|| ||3||Design the Daughter board||8h|| ||4||Design the packaging||16h|| ||5||Write firmware||40h|| ||6||Assemble units||8h|| 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''. ||'''ID'''||'''Summary'''||'''Estimate'''||'''Predecessors'''|| ||1||Design the Schematic||24h|||| ||2||Design the Motherboard||8h||1|| ||3||Design the Daughter board||8h||1|| ||4||Design the packaging||16h|||| ||5||Write firmware||40h||1|| ||6||Assemble units||8h||2,3,4,5|| 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. ||'''ID'''||'''Summary'''||'''Work'''||'''Predecessors'''||'''Due'''||'''Start'''|| ||1||Design the Schematic||24h||||May 5, 2009||May 3, 2009|| ||2||Design the Motherboard||8h||1||May 10, 2009||May 10, 2009|| ||3||Design the Daughter board||8h||1||May 10, 2009||May 10, 2009|| ||4||Design the packaging||16h||||May 10, 2009||May 9, 2009|| ||5||Write firmware||40h||1||May 10, 2009||May 6, 2009|| ||6||Assemble units||8h||2,3,4,5||May 11, 2009||May 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. ||'''ID'''||'''Summary'''||'''Work'''||'''Predecessors'''||'''Resource'''||'''Due'''||'''Start'''|| ||1||Design the Schematic||24h||||||May 5, 2009||May 3, 2009|| ||2||Design the Motherboard||8h||1||Monty||May 9, 2009||May 9, 2009|| ||3||Design the Daughter board||8h||1||Monty||May 10, 2009||May 10, 2009|| ||4||Design the packaging||16h||||||May 10, 2009||May 9, 2009|| ||5||Write firmware||40h||1||||May 10, 2009||May 6, 2009|| ||6||Assemble units||8h||2,3,4,5||||May 11, 2009||May 11, 2009|| With 50% applied to each task, the board designs are worked on in parallel but take twice as long. ||'''ID'''||'''Summary'''||'''Work'''||'''Predecessors'''||'''Resource'''||'''Due'''||'''Start'''|| ||1||Design the Schematic||24h||||||May 5, 2009||May 3, 2009|| ||2||Design the Motherboard||8h||1||Monty (50%)||May 9, 2009||May 10, 2009|| ||3||Design the Daughter board||8h||1||Monty (50%)||May 9, 2009||May 10, 2009|| ||4||Design the packaging||16h||||||May 10, 2009||May 9, 2009|| ||5||Write firmware||40h||1||||May 10, 2009||May 6, 2009|| ||6||Assemble units||8h||2,3,4,5||||May 11, 2009||May 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. ||'''ID'''||'''Summary'''||'''Work'''||'''Predecessors'''||'''Resource'''||'''Due'''||'''Start'''|| ||1||Design the Schematic||24h||||||May 1, 2009||April 29, 2009|| ||2||Design the Motherboard||8h||1||Monty||May 7, 2009||May 7, 2009|| ||3||Design the Daughter board||8h||1||Monty||May 8, 2009||May 8, 2009|| ||4||Design the packaging||16h||||||May 8, 2009||May 7, 2009|| ||5||Write firmware||40h||1||||May 8, 2009||May 4, 2009|| ||6||Assemble units||8h||2,3,4,5||||May 11, 2009||May 11, 2009|| Microsoft Project supports many calendars in a project. Roughly speaking, the project calendar has working hours per day and working days per week (e.g., 9-5, Monday-Friday) and resource calendars can override that (e.g., Ethan is on vacation the August 15-19). == 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. [[Image(simple-gantt.PNG)]] 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. [[Image(gantt-progress.PNG)]] === 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. [[Image(gantt-resource.PNG)]] 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''. [[Image(critical-path.PNG)]] 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. [[Image(roll-up-1.PNG)]] [[Image(roll-up-2.PNG)]] === Progress Line === A Gantt chart often includes a "today" line to show where a project stands. [[Image(progress-line.PNG)]] 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. [[Image(slack.PNG)]] 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. [[Image(resource-utilization.PNG)]] == 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 Core == There are a couple of tickets ([http://trac.edgewall.org/ticket/31 31], [http://trac.edgewall.org/ticket/886 886]) and some discussion about ticket [http://trac.edgewall.org/wiki/TracDev/Proposals/TicketLinks linking] and [http://trac.edgewall.org/wiki/TracDev/Proposals/TracRelations relationships] in the Trac core. We may be able to get a lot done while we wait for that work to come to fruition. == Trac Plugins == === Gantt Charts === Trac has several Gantt chart plugins: [wiki:TracJsGanttPlugin TracJsGantt]:: A data-driven plugin which presents ticket status in a display much like Microsoft Project's WBS format. [wiki:FlashGanttPlugin FlashGantt]:: seems complex and stale. [wiki:GanttChartPlugin GanttChart]:: is a very simple Gantt chart implemented in a formatted text region. It is manually edited, not data-driven. [http://willbarton.com/code/tracgantt/ 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. GanttCalendarPlugin:: provides nice views of tickets. It is noted as "not stable". It uses "Completed [%]" (quantized to 5% increments), "Start" (YYYY/MM/DD), and "End" (YYYY/MM/DD) custom fields. It seems to include an administrative calendar, perhaps where holidays and such can be recorded. Pale Purple's [http://virtual-planner.palepurple.co.uk/ Virtual Planner] is an alternative visualization tool. === Dependencies === MasterTicketsPlugin supports FS dependency (but calls it ''blocks'' and ''blocked by''). The [http://trac.edgewall.org/wiki/SubTickets SubTickets] page talks about adding composition type dependency (parent/child relationships). SubticketsPlugin and ChildTicketsPlugin provide composition type dependency that can be used for WBS. The [http://trac-hacks.org/wiki/TracDependencyPlugin Trac Dependency plugin] was created in August 2009 and shows promise as a more complete and flexible solution than upgrading !MasterTickets. === Time and Scheduling === 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. DateFieldPlugin has some helpful wrappers around custom fields to validate them as dates. 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 TeamCalendarPlugin keeps track of user availability. === Data Exchange === The TicketImportPlugin can import tasks exported from Microsoft Project as a CSV file. There is a [ticket:7812 patch] to import dependencies so Microsoft Project can feed !MasterTickets. 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 [http://www.viewpath.net/Website/ganttcharts/googlegadget.aspx Google Gadget Gantt chart] which might be leveraged for the visualizations once the data is made available (e.g., using [https://opensvn.csie.org/traccgi/swlcu/wiki/En/Devel/TracGViz/ TracGViz plugin]). That widget is developed and maintained by a company (i.e., [http://www.viewpath.net/ 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. [http://www.simpleprojectmanagement.com/planner/home.html Planner] is an open source project management program. GNOME Office once included [http://wiki.linuxquestions.org/wiki/MrProject MrProject] (which may be another version of Planner). The news of !MrProject's death may be exaggerated; [http://live.gnome.org/Planner GNOME Live!] lists a 0.14.4 release in April 2009. [http://www.dotproject.net/index.php 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 [http://en.wikipedia.org/wiki/DotProject 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 [http://www.web2project.net/ Web2Project]. I reviewed [http://www.smashingmagazine.com/2008/11/13/15-useful-project-management-tools/ 15 Useful Project Management Tools]. It mentions Trac. None of the tools obviously have scheduling in them. [http://www.atlassian.com/software/jira/ Jira] sounded promising but doesn't seem to have scheduling. Their chart examples don't include a Gantt. [http://www.basecamphq.com/ 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: * [http://www.openworkbench.org/ Open Workbench] is a Microsoft Windows-based application not suitable for integration with a web-based tool like Trac. * [http://openproj.org/openproj OpenProj] also appears to be a desktop application. However, their [http://openproj.org/wiki/index.php/Available_Views views] page may guide development of PM output for Trac. * [http://www.ganttproject.biz/ 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. * Note that there is a [ticket:7812 patch] to TicketImportPlugin that imports WBS relationships. It is known to work with !MasterTickets and Subtickets plugins and may work with others. * !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 or configuration or derive them from history for experience-based scheduling (rather than have rules to use default values for tickets with tiny estimates). * A simple Gantt chart can be implemented to show dependencies and schedule based on manually-entered due dates. * TracJsGanttPlugin provides such a display. * [#DependencyandDuration Basic scheduling] (ignoring resource conflicts and availability) can be provided based on * core Trac tickets, * !MasterTickets, * !TimingAndEstimation, and * custom fields to hold assigned and calculated dates[[br]]A new Scheduling plugin can build on these plugins (or an interface that hides them), and create an as-late-as-possible schedule working back from a milestone. * We need a resource calendar plugin to allow describing resources and their availability. This should support notions of normal work hours, holidays, etc. * 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 = To maximize the flexibility in mixing and matching plugins to provide features for project management, I propose to leverage Trac's [http://trac.edgewall.org/wiki/TracDev/ComponentArchitecture Component Architecture] to specify a number of interfaces in a `tracpm` namespace. In this way a Gantt chart, a workload chart, a task scheduler, etc. can all use the `tracpm` interfaces regardless of whether, for example, Subtickets or Childtickets (or even ticket decomposition in a future Trac core) are used to represent parent/child relationships. == 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 !TimingAndEstimation 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, user-specified constraint on when the task must start (a date) assignedFinish:: An explicit, user-specified 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 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; though both could be set if ''percentEffort'' was allowed to be computed). 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: '''Used to have `startFromFinish()` and `finishFromState()` here but those are likely scheduler functions. I think what we need here is `hoursAvailable(resource, date)` which a scheduler can call on each date and subtract the result from the work it is trying to schedule.''' == 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, the resources assigned to those tasks, and the resource availability. Where the tasks do not have ''assignedStart'' or ''assignedFinish'', the computed schedule prioritizes tasks to keep resources at or below availability. 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 predecessors 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[[br]]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, [http://en.wikipedia.org/wiki/Operations_research OR]) (cf. [http://books.google.com/books?id=pHetPl2LOYgC&lpg=PA153&ots=rKKZ7Lv2On&dq=pslib%20project%20scheduling&pg=PP1 Project Scheduling]) and has been established to be [http://en.wikipedia.org/wiki/NP-hard 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. [http://129.187.106.231/psplib/ PSPLIB] provides a standard set of program scheduling problems to test the various algorithms performance against one another and the optimal solution. Kolisch and Hartmann[[FootNote(Kolisch, R. & Hartmann S. "Experimental investigation of heuristics for resource-constrained project scheduling: An update" European Journal of Operational Research 174 (2006): 23-27.)]] tested dozens of algorithms and variations using the PSPLIB data. ---- [http://portal.acm.org/citation.cfm?id=1162708.1163079&coll=ACM&dl=ACM&CFID=42032534&CFTOKEN=18432343 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 ... [http://portal.acm.org/citation.cfm?id=1351542.1351869&coll=ACM&dl=ACM&CFID=42032534&CFTOKEN=18432343 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 ... [http://portal.acm.org/citation.cfm?id=324898.324933&coll=ACM&dl=ACM&CFID=42032534&CFTOKEN=18432343 Activity scheduling in the dynamic multi-project setting: choosing heuristics through deterministic simulation] Robert C. Ash December 1999 And others at [http://portal.acm.org/results.cfm?coll=ACM&dl=ACM&CFID=42032534&CFTOKEN=18432343 ACM.org]. == Other `tracpm` functions == I'm not sure what interface to put these in but some other functions that the API should hide are: * Finding predecessors (immediate and indirect) * Finding successors (immediate and indirect) * Finding descendants (children and further generations) * Finding ancestors (immediate and further generations) * Finding all related tickets (predecessors, successors, descendants, ancestors, and possibly those that share a resource) == 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. [#Contributors 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 [[FootNote]] ---- = 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. 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. 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.