Risk-Driven Model For Task Allocation

You can download the original paper from : Risk-Driven Model For Task Allocation

You can download the presentation from : Risk-Driven Model For Task Allocation 

     In the GSD scenario, the stakes are comparatively high and most of the risks or potential benefits are directly depended on the way the task are assigned to the different sites. There is a need to assess the risks involved in order to get the work done effectively. This post is a summary of the paper "A Risk-driven Model for Work Allocation in Global Software Development Projects" by Ansgar Lamersdorf, Jürgen Münch and Alicia Fernández-del Viso Torre.

     There are a lot of risks involved in GSD scenario  And most of them occur because of inefficient task allocation. Take for example the problem of language, culture, time zone, abilities etc etc. And one of root problem for all these problems is the task allocation. If we can address the task allocation problem effectively, if we can find a metric or a model to suggest task allocation possibilities to the decision makers, the risks involved can be reduced by a huge factor.

     The authors have developed a model that does exactly this. They made an Integrated Assignment Model to address this problem. The problem of task allocation is to find the best match between elements of two sets (as shown in the fig below).

1. A set of tasks that together form the software development project. Ex. a needed role, responsibility for a certain part of the product, a process step etc.
2. A set of sites (locations) that together form the available resources.

      Now the model has to identify the different task allocation possibilities and then suggest the best assignment. For doing this, of course, it has to consider an organization specific set of influencing factors like time complexity, experience, task coupling, time zone differences etc.). And we also need to consider the impact of these factors on the project goals.

The authors formulated the following goals for their work:

1. Document the decision process
2. Suggest work assignments
3. Evaluate the consequences of different alternatives.

MODEL:

      Based on a division of the project and resources into tasks and sites and a subsequent characterization of the tasks and sites, the model is able to provide a weighted list of assignment suggestions and to evaluate every alternative with respect to the expected project risks. The figure below gives an overview of the model input and output.

The model consists of three main sub models

1. Casual Model
2. Stochastic Assignment Model
3. Risk Identification Model

Casual Model:

The casual model stores the organization-specific, relevant influencing factors and their impact on project goals as shown in the figure. The weights of each relationship is calculated based on the past experiences in other GSD projects.

Stochastic Assignment Model:

The work of this sub-model is to do the following:

1. The causal model is transformed into Bayesian networks.
2. Using the characterization of the project and the resources, the impact of every possible assignment on project goals is inferred in the Bayesian networks. The impact is aggregated based on weighted project goals and stored as probabilistic distributions for the cost functions needed in the task assignment algorithm.
3. In a large number (e.g., 1000) of runs, the cost functions are instantiated based on the probabilistic distributions and the assignment algorithm from distributed systems is executed.
4. For every run, the returned optimal assignment is stored. All returned assignments are then ordered by their number of appearance (i.e., the number of runs in which the assignments were returned). This is
5. finally presented as a weighted list of assignment suggestions.

Risk Identification Model:

The risk identification model is able to predict GSD-related risks for a given project and work allocation. Therefore, it can be used for analyzing and comparing the different assignment alternatives suggested by the application of the stochastic assignment model. The model does this by transferring lessons learned into a set of semi-formal logical rules.

Does the model work..?

Yes, the model seems to give reasonable suggestions. The test of its results were compared to those of managers and it was shown that the model in fact works most of the time.