Three-layer model to be built. 

Context in source publication

Context 1

... first one deals with the modelling paradigm. Before years 2000, IS capacity planning models were essentially based on queuing network models (QNM). Later on, those models were adapted to the increasing complexity of the industrial IS, for instance client/server architecture [ 18] or large-scaled distributed systems [8]. These models are always abstract and built with a top-down approach [13] and have to be considered as “analytical” models, in the sense of Le Moigne [14]. However, from the years 2000’s a change in the modelling approach can be underlined, characterized by an increasing use of “systemic” models. To handle the increasing complexity of IS systems from a capacity planning point of view, the question is no more to represent how they are constituted but to understand how they are working (knowledge in action) [3, 14]. This new mindset is paying more attention to datacollection and historical analysis. Data are collected through simulation [7] or historical data storage [1]. We can also distinguish two trends in the “systemic” models: white boxes, which are trying to explain real interactions [1, 7] and black boxes, which are looking for overall patterns that could be identified anywhere within the overall system [12]. However, such “systemic” models have also their limits. A side effect of this new paradigm can be illustrated by Allspaw’s motto: “ Measure, measure, measure ” [1]. Indeed, industrials are now gathering huge volumes of IS activity data, which tends to increase the complexity of the capacity planner’s work : how to extract useful information from this amount of information? The second key-point to be addressed is the link between production and IS, notably through the use of business workload variables within IS capacity planning models. In the scope of the analytical models, before the 90’s, business activity and IS resource models were two different topics in the capacity planning literature. After this period, we can observe the emergence of more global models, which are gathering workload, including “ natural forecast units” – e.i. business workload -, and IS resources into one single model [2]. [19] also proposes to integrate complex workloads (not in natural forecast units, however), by decomposing it into basic components, within a tree structure. Though, as these proposals are done in the scope of QNM, the workload is finally characterized as a mono-dimensional technical information, as an input of the QNM. In the scope of the “systemic” models, business information are analysed at the same level as any other information. However, here again some limits have to be underscored: in typical capacity planning models, business workload and IS resources capacity interactions are never formally quantified, neither in analytical nor systemic models. The objective of the method developed in collaboration with ST-Microelecronics is to go beyond such limits of the current scientific literature : - On the one side, we propose to use a systemic model, with a reduced amount of information to be managed when maintaining and updating this model. - On the other side, we propose to formalize and quantify correctly the links between production system variables (like business workload) and the IS behaviour. Thus, the following sections propose an alternative use of “systemic” models, requiring to build a specific IS model, which will be later used to test complex “what-if” scenarios, with business variables as inputs to represent business changes, and with a quantification of change impacts on the IS performance. As mentionned earlier, the current paper is limited to discuss the method proposed to build the model. The purpose is to configure a rigorous method, aiming at building a multi-layer model, linking the IS architecture and the production system. When defining this method, some specifications have to be considered: our approach is systemic, not analytic • the model has to be a white box one, highlighting concrete interactions between business activity and IS resources • the modelling process is starting from scratch, without any knowledge on the studied IS • only the use of the IS resources is handled, the system performance is out of scope • interactions between business workload and resource capacity have to be quantified • the business workload is complex: potentially multi-dimensional and multi-levels (tree structure) • no simulation for datacollection: we are following a secondary analysis process [9] • we are building our model by firstly analyzing the resources usage, and then trying to understand how it could be explained by business activities variations [5] Additionally, this method will mix quantitative and qualitative approaches, as explained earlier. Referring to figure 1, the objective is to build a model linking business activity to IS resource usage, through a bottom up approach aiming at identifying predictive links between low level variables (variables of the “resource model” level) and upper level variables (“functional model”, then “business model” variables). Confronted with several hundreds of potential variables (notably in the functional level), we follow a procedure with two main phases: an exploratory phase aimed at selecting relevant modeling variables, then an exploratory phase to identify quantified relationships among variables. We explain below these two steps, with an example applied to the creation of a model linking “resource model” variables with “functional model” variables. As underlined in figure 2, the exploratory and explanatory phases are combined via an iterative process. The current section provides a general overview, and the two phases of the method will be illustrated in section ...