Research
I completed a Masters Degree in Engineering specialising in Project Management at the Univestiy of Pretoria. I graduated in April 2007. As part of my degree I completed a research project on critical chain project scheduling and more specifically on buffer sizing for critical chain project scheduling. I was required to compile a research report as well as a journal article. Prof. Herman Steyn of the University of Pretoria co-authored the journal article. The article will appear in a future edition of the South African Journal of Industrial Engineering (SAJIE). I am not sure of the rules and regulations pertaining to the publishing of journal articles but just in case SAJIE has some sort of exclusive right to the article I won't place it here. You can however read my full research report if you are interested (or if you suffer from insomnia.) The full report is basically the 144 page version of the 15 page journal article. At least you won't miss out on any of the detail! The abstract from the full research report is presented below.
You can dowload the full report in pdf format here
Abstract
This research investigates various aspects of buffer sizing for project and feeding buffers for the critical chain project management method. (CCPM) An historical background to project scheduling methods is presented. Charting, network scheduling, critical path, PERT and Monte Carlo simulation are addressed. CCPM is discussed in detail.
A review of the relevant literature for buffer sizing and buffer management is presented. The review highlights aggregation of risk and addresses theoretical aspects of task properties and combinations of tasks and their impact on buffer consumption. Seven buffer sizing methods from current literature are discussed. They are: The cut and paste method (C&PM); the square root of the sum of squares method (SSQ); the SSQ with bias method; the adaptive procedure with resource tightness (APRT); the adaptive procedure with density (APD); there is also a model which proposes buffer sizing on the basis of relative dispersion of the chain and a desired level of safety and the final buffer sizing method is one which sets out to address systemic errors in scheduling method and sets a lower bound to the buffer size to account for bias factors. Finally, the literature review address buffer management methods and highlights fixed intervention approaches compared to relative and trend based intervention approaches.
Three core models are developed, presented and analysed. The first investigates the relationship between chain mean duration as well as standard deviation of chain mean duration and buffer consumption. The second model investigates the impact of multiple parallel chains on merged activity start time. The third model compares the performance of the C&PM and the SSQ method of buffer sizing for cases with and without bias.
The key findings are: In the absence of bias buffer consumption is dependant only on chain or task standard deviation and is completely independent of chain mean duration. For certain classes of bias buffer consumption remains independent of chain mean duration but for others buffer consumption is a function of chain mean duration. Buffer consumption increases with increasing numbers of parallel paths and with increasing standard deviation of the paths. Buffer consumption is independent of parallel path mean durations. Initially the addition of parallel paths has a marked impact on buffer consumption but as the number of paths increases the addition of further parallel paths has a smaller relative impact. In the absence of bias, except for very short chains of tasks where the task variance is extreme, buffers sized using the C&PM are excessively large and uncompetitive. Even with bias C&PM buffers appear to be too large but this should be viewed with caution depending on which forms of bias are present. SSQ buffers perform very well in the absence of bias but exhibit unacceptably poor performance for optimistic estimate bias and gold plating bias. A new approach to buffer sizing is proposed whereby a fixed buffer portion is determined on the basis of a database of previous project performance in the organization concerned and a variable buffer portion is calculated on the basis of task variability.
Keywords: Critical chain, buffer sizing, Monte Carlo Simulation, bias, buffer management.