Modelling processes to scale

Tasks in process models come in varying sizes. There is a key distinction that divides tasks in process models into two classes: is this task at the "plannable" or "sub-plannable" level. Plannable tasks appear on plans, Gantt charts, burndowns and crucially time-sheets. They are the kind of tasks that take from several hours to days or even weeks. Plannable tasks are the trackable planned, estimated and costed elements in xProcess. However in some processes we wish to define (and monitor) tasks at the sub-plannable level. For example if my tasks is to distil a certain compound from ingredients (in a chemical process), the steps in the process may need to separately defined and tracked to ensure quality compliance. xProcess provides a mechanism to support this in the "Gateway Type". While these steps form only a simple sequence in v2, we've recently been experimenting with a richer model for gateways that will allow the same kind of forms - selections, collections, iterations and sequences - that are supported in plannable tasks through the composite structures. Together these facilities make a continuum of process modelling from from large to small scale.

Better Faster Audit

I've recently been asked to outline the structure for our process audit service offering at xProcess Europe. The product is called the "Better Software Faster Process Audit" and is based on the structure of Dan Haywood's and my book of the same name(Better Software Faster). It covers 7 crucial aspects of any software development process:

• The last step - Build and Deploy
  
• The first step - Modelling and design
• The stakeholder step - Requirements
• The controlling step - Management
  
• The continuous step - Quality measurement
• The micro step - Development
• The macro step - Architecture

Starting with the last step usually throws people, but given one of the key criteria for improvement is agility, it makes sense to start with seeing how close a team is to fully automating their build and test suite. There's no excuse for not fixing this area if you're using languages like Java or C# but with some older languages it can be tough to provide continuous integration for every change made by a developer.

Quality Measurement is another related area which is sometimes overlooked. Every team has some means to assess quality. For some teams "release the product and monitor the bug reports" seems to be their preferred way of working. Not recommended!

EVM and Agile Processes

I am presenting a paper on EVM and xProcess next week at the 2nd EVM Conference in Birmingham, UK. I’d love to see you at the conference but if you’re not able to make it and would like a copy of the paper you can download it now from the xProcess Europe or Ivis web sites. Here's the abstract...

EVM and Agile Processes – an investigation of applicability and benefits

Earned Value Management (EVM) is a technique for understanding the degree to which a project is following a plan. A prerequisite for EVM as traditionally applied is therefore a stable baseline plan with a fixed scope. Agile processes for software development by contrast are designed to be responsive to a changing environment and enable discovery of requirements through their lifecycle. Agile plans are rarely stable or of fixed scope. However both approaches share the goal of producing plans that provide stakeholders with immediate feedback on the daily progress of projects and they share common needs in understanding and reporting progress and productivity. In addition the fact that agile processes map requirements to plannable tasks means that requirements may change without invalidating the baseline for progress reporting. This paper reports on work to develop tool support based on the xProcess process and planning product to apply EVM to agile processes. It shows how the metrics for schedule and cost efficiency can be modified for compatibility with agile approaches, and how they help effective management of agile projects, particularly when supplemented with tool support for project forecasting and monitoring.

Calculating team velocity

Why is it that agile methods use timeboxes containing broadly similar sets of activities - as opposed to phases favoured by more traditional methods, which contain different types of activity in each phase through the lifecycle? A key reason is that stakeholders in the project get quicker and more effective feedback on the progress and direction of the project. As a consequence planners can judge more accurately the effectiveness of the project and what it is likely to achieve over its full duration. One of the most useful metrics planners get from each timebox is the velocity of the team. Scrum projects usually use the velocity from previous sprints to forecast their likely progress - a technique often referred to as "yesterday's weather" since it effectively assumes the next period will be broadly the same as the previous. In this article we discuss what velocity is, how to calculate it in xProcess, and how to go beyond "yesterday's weather" by calculating and using simple productivity for improved forecasting.

Velocity is defined as the amount of required work delivered per time period. In the Scrum process for example, required work is called the Backlog, and its component parts, the Backlog Items, are individually sized and prioritised. The size of an item is not the same as the effort required to complete it, though size and effort are related and may - once the velocity of the team is known - be derived one from the other (see here for further discussion of size and effort). The size of backlog items is usually estimated in a team-specific measure called points (sometimes called Story Points, Feature Points or, if related to the amount of work a team member could carry out in a full uninterrupted day, Ideal Days). The size estimates therefore really only give a relative size for the backlog items. However once a timebox has been completed - in Scrum timeboxes are called Sprints - the team has a measure of how many points they were able to complete and so have the first reading for velocity in Points per Sprint.

When using xProcess for Scrum you can see the velocity of completed Sprints from the burndown chart. For example in the Sprint shown below you can see that a total of 91 points were completed, giving the team a velocity of 91 points per sprint.

Using the principle of yesterday's weather it might be natural for the team to commit to a similarly sized set of backlog items in the next sprint. Before doing so they might be wise to look at all the history on the project. Here are the burndown charts for the previous two months, first Sprint 02 and then Sprint 01...
As we can see from the respective velocities of 40 points per sprint and 73 points per sprint, velocity is not always constant and, if we look at the reasons behind this, we may be able to forecast the velocity for the next sprint more effectively. Why did Sprint 02 for example do so badly?

Well one clue is in the dates for this Sprint. This won't be the only team to have discovered their productivity was lower between mid-December and mid-January! All the team were on holiday for at least some of the Sprint, and when they were in they probably discovered someone they needed to speak to was away, or that there were other activities they had to attend to which were not related to the sprint backlog. Another factor was that this sprint had a number of items that had been worked on but not completed at the end of the sprint. Since they are not finished they cannot be counted in the size of work "done". Backlog items like this do not necessarily get done in the following sprint since other priorities may be introduced to the team at that point. However if they are carried forward (as they were in this case), there will be considerably less work left to complete in the following sprint. This at least partially explains why Sprint 03 was such a productive one.

So what velocity should the team predict for Sprint 04? This is where other information from xProcess comes in very handy in the planning process. xProcess allows team members to record when they were working on overhead tasks (like management meetings, email and admin, training courses and office parties) as opposed to backlog items themselves. xProcess also give us information about the amount of time spent on backlog items that were not completed in the sprint.

The key metrics we want to derive in order to better predict the next sprint are these:

• simple productivity (size of work done in points per ideal person-day)
• overhead percentage (what proportion of people's time is applied to non-backlog-item tasks)
• availability (when are people available to the project)

We'll leave discussion of overhead tasks and availability to another occasion, though they can be set simply in xProcess. The simple productivity measure of Size/Effort can be calculated automatically by xProcess. I say simple productivity measure since, as Putnam and Myers point out, it is more complex to derive a true process productivity for a project, taking into account that size and effort do not follow a linear relationship when the duration under consideration varies. This non-linearity can be broadly discounted however if the duration of the sprint does not vary. (One of the reasons why agile methods recommend fixed durations for timeboxes is to ensure that feedback from one timebox is applicable to the next.)

Simple productivity for a given period (a sprint for example or the whole project) takes into account all the non-overhead tasks booked to during the period. It also takes into account tasks which were incomplete at the start and end of the period. Once the simple productivity factor has been calculated for previous sprint (or for the project so far), the value can be use to set new effort estimates based on size using the Set effort to match size UI Action described in a previous article.

The simple productivity currently being used on this project is 1.0 - in other words a backlog item estimated as size 1 will be scheduled with an estimated required effort of 1.0 ideal person-days. Using this factor the current state of the burndown chart for Sprint 04 is as shown below.


However using the history from the previous 3 sprints, xProcess can calculate the actual productivity in points per ideal person-day. In this case it was found to be 0.91 - in other words historically a task of size 1 has taken on average about 8.8 person-hours to complete. Using the Set effort to match size action, we can quickly adjust the estimates of all the open tasks in the project to reflect this factor. Once all this is applied (see burndown chart below), the state of Sprint 04 now reflects the true situation that certain tasks are at higher risk of not completing.

Developing your own workflow monitor

Just occasionally I'm tempted to post a really techie article that will probably only be of interest to hard-core xProcess enthusiasts! This is one of those occasions. I've written elsewhere about some of the aspects of workflow support in xProcess (see for example the article on State Diagram support which explains how to define triggers for internal events graphically, and "From tracking to planning and back" which discusses how tracker products can be linked to xProcess via workflow). This article though will go a bit further and give you some hints about how the Workflow Server works and how you can link in your own programs to detect workflow events and trigger actions on them.

The Workflow Server sits in a Tomcat instance watching the outside world (and the inside world of the xProcess Data Source) through a set of monitors. When these monitors discover events that might be of interest, they tell the Workflow Server which checks whether any object in the xProcess Data Source has registered an interest in that type of event. If it has it invokes the defined action(s) for that event, potentially committing changes to the xProcess data or indeed to external systems. Several monitors are delivered with the xProcess Workflow Server. There's the DataSourceMonitor itself which monitors state transitions within the Data Source.There are also monitors for the Flyspray bug tracking system, for CollabNet's Project Tracker and a general purpose email monitor. However if you want to integrate external systems with xProcess this is exactly the way to do it - write your own monitor. In spite of my introduction to this blog article, this is quite straightforward to do. There are 3 key elements:

1. A monitor class (inheriting from com.ivis.xprocess.workflowserver.monitors.MonitorImpl)
2. Methods for this class which initialize it, find the events
   
3. [optionally] An event class (inheriting from com.ivis.xprocess.workflowserver.ExternalEvent) and possibly useful methods that may be called when the event is triggered.

Here (below) is a code snippet for a sample monitor which show the key methods init(), which is called when the Workflow Server starts up, and findEvents(), which is called periodically by the server in order for this monitor to query its external system and provide the events to the server.

public class SampleSystemMonitor extends MonitorImpl {
SortedSet events;
public static String EXTERNAL_SYSTEM_NAME = "My External System";
public static String EXTERNAL_EVENT_NAME = "My event name";

public String getSystemName() {
return EXTERNAL_SYSTEM_NAME;
}

public String[] getEventNames() {
return new String[] { EXTERNAL_EVENT_NAME };
}

@Override
public void init(DataSource dataSource) throws WorkflowException {
super.init(dataSource);
// TODO initialize connection with external system
}

@Override
protected SortedSet findEvents() throws WorkflowException {
events = new TreeSet();
// TODO query external system for new events, e.g.
// loop
ExternalEvent xevent = new SampleSystemEvent(new Date(), EXTERNAL_EVENT_NAME);
Map map = xevent.getMap();
//TODO set map values so they can be referenced in the triggered action, e.g. ...
map.put("name", "New Task");
events.add(xevent);
// end loop
return events;
}
}

The monitors are singleton objects and are instantiated within the server by Spring. So having written your monitor you can inform the Workflow Server that it should be called simply by modifying the xprocess.xml in the Tomcat WEB-INF directory (following the pattern of other monitors). Specific objects in the xProcess Data Source can respond to the events and trigger specific actions through their "WorkflowPackage" - which is declared in the process definition.

Set effort to match size

A previous entry provided the script required to define an action shortcut for setting the Size of tasks based on existing estimates for the effort required (see Set size to match effort). In some cases the inverse of this operation - Set effort to match size - is the useful operation. Here's the code you need for that one. Once again you should make it a "UI Action" by setting this field to True so that it appears on the right-click menu when you select a task or set of tasks.

Action Name: Set effort to match size
Expression:
#factor = getProject().getFloatProperty( "CurrentProductivity_pointsPerIdealPersonDay"),
#factor = (#factor < 0.01) ? 1.0 : #factor,
#size = getSize(),
setBest(#size/#factor*240.0),
setMostLikely(#size/#factor*480.0),
setWorst(#size/#factor*720.0)
Applicable to: Task
UI Action: True
Parameters: (none)

Note that the operation uses a factor to scale the mapping between size points and effort. If this factor is 1.0 the mapping is 8 hours effort (480 minutes) per size point.

Perhaps - like it's inverse operation, this will one day appear in the preconfigured operations of the Simple Process. Note that is very similar to the Set estimates action which is delivered in this process. However that operation is designed to take Size as a parameter (usually during the instantiation of a pattern) and so it is not suitable to make that one into a UI Action for operating directly on existing tasks.

Traceability through a simple metamodel

When people start defining processes with total traceability between every aspect, the sight is often not very pretty! When traceability adds multiple layers of complexity it rarely pays for itself in terms of improved quality or maintainability.

This is one of the reason why we defined what we termed the "minimum metamodel" in the Better Software Faster book on agile development. The diagram shown here is this model in its simplest form. The questions traceability should answer are:

• How is this requirement tested?
• How is this requirement implemented?
• How is this part of the design tested?
• Why is this part of the design needed?

It's a useful starting point when defining the artifacts required in a software development process, and the references they should contain to enable these questions to be answered.