Efficient Use of Space Over Time Deployment of the MoreSpace Tool Štefan Emrich Dietmar Wiegand Felix Breitenecker Marijana Srećković Alexandra Kovacs Shabnam Tauböck Martin Bruckner Benjamin Rozsenich Salah Alkilani Niki Popper Vienna University of Technology, Austria, (e-mail: stefan.emrich@tuwien.ac.at). Vienna University of Technology, Austria dwh GmbH - simulation services, Vienna, Austria Abstract: As proposed in a study conducted by Wiegand et al. (2007) at the ETH Zürich, educational facilities hold a high potential yield for improvement of room utilization. The goal of the project MoreSpace at Vienna University of Technology (TU Vienna) is to develop a new modeling approach which helps to use space efficiently. For a successful deployment of this model its requirements (preconditions, input-data, dissemination, etc.) have to be met by the peripheral system. This paper covers the methods applied for analyses of the model and the system, which enable model integration as well as the insights gained during the course of this process. It further describes a deployment matrix puts the mode of operation into context with met preconditions and the required depth of system integration. This allows it to estimate whether a model can be deployed as intended or not; with alternatives being either a transformation of the system, reformulation of the question(s) towards the model or in the worst case abortion of the deployment process. Keywords: Business process engineering, System analysis, System integration, Data processing, Data Visualization 1.1 Overview 1. INTRODUCTION While the overall goal of the simulation is to increase the utilization of (lecture) rooms, there are several approaches to reach it. Depending on the questions formulated towards the simulation model, differing requirements are implied which influence the mode of operation, necessities towards (input) data, model preconditions, depth of system-integration, quality management, etc. 1.2 Modes of Operation Regardless of the precise circumstances of the simulation project, the following three general operational modes can be distinguished based on the necessary integration within the peripheral system: (1) Day-to-day operations (recurring) The simulation model is being used on a regular (e.g. hourly, daily, weekly) basis to generate information which is immediately used by the surrounding system for day-to-day operations. Example: Simulate & evaluate room booking strategies and chose most suitable for current booking period. (2) Strategic planning System assessment (recurring) The simulation is used frequently (or even on a regular basis) with the results generated being of an informative nature, used for strategic planning. Example: Simulation runs based on historical data to test & compare different management strategies. (3) Consulting purposes (non-recurring) Projects to answer a single question (e.g. for decission making). Often such a simulation will be used only once and without the need for deep integration within the systems processes. Example: Consulting jobs to evaluate and decide on best factory layout for a production process. The necessary depth of integration of the model within the surrounding system is closely connected to the operational mode. While in most cases it would be plainly uneconomic to deeply integrate a model intended for a mode 3 utilization, as the dissemination of the outcome is punctual (temporally as well as stakeholder-wise). A mode 1 model will be unlikely to function properly if it is not perfectly coordinated with the peripheral system (e.g. assure constant acquisition of qualitative up-to-date input data). The aspect of system integration will be covered in more detail in a later section. 2. DEPLOYMENT SCENARIOS In order to increase the room utilization three scenarios have been developed for the deployment of the MoreSpacetool. These scenarios address each of the before mentioned operational modes, providing an example to study for each mode.
In order to avoid misunderstandings, following two definitions have to be coined: the room utilization and a room s degree of exploitation. While the room utilization is defined intuitively as the fraction of time during which the room is used over the total time in which the room is available, the degree of exploitation goes one step further. It is defined as the ratio with which a room s capacity is being used. It is calculated as the fraction of people inside divided by the capacity of a room. 2.1 Operational Deployment Supporting the Booking System The first scenario applies to the core booking process at TU Vienna. In this case the MoreSpace tool has to simulate the booking (assignment of an adequate room to every authorized room request) of all courses. Afterwards the model simulates the user behavior (students with their individual curricula) in order to verify the previously established bookings and to evaluate the degree of exploitation of the rooms 1. This strategy has to be applied in the phases during which room requests are issued. Even though the main part of all courses is scheduled prior to the start of teaching (beginning of the semester), a substantial part of requests will occur during the semester. 2.2 Strategic Deployment Monitoring of Peripheral System The second deployment scenario is aimed towards strategic management and the monitoring of the system. For this purpose the MoreSpace tool is to be used to analyze the system with respect to various questions such as: what is the university s limit of student numbers?, or could a different curriculum have improved last semesters room utilization?. As visible from the types of questions, such an application will have to rely on historic data eventually enhanced by projections and forecasts. Regardless whether the simulation is directed backwards or forward. 2.3 Deployment for Consulting Purposes Decision Support By definition, a questions which calls for a solution relying on external simulation will be independent of models that are already deployed. Thus the present scenario is not a perfect representation of this third operational mode, but more of a gray area between strategic deployment and consulting. This scenario covers reconstruction phases, with the associated shut down of rooms. To find an acceptable sequence for the shut down of rooms, testing would be conducted with the MoreSpace tool prior to reconstruction. After such a sequence is identified the question can be regarded as solved without the necessity for further simulation runs. 2.4 Structure of Simulation Model Based on the specified deployment scenarios the model s structure can be derived. While the aim of the model is to increase the efficiency of room usage, the core necessity 1 The room utilization can be derived already from the simulation s first part. of the simulation is to satisfy all bookings requests and important user demands 2. Thus the courses and the rooms pose the central elements of the system. The third main element are the students the users within the system. A possible and elegant mapping of these entities is to regard the rooms as servers and the courses as products which are processed by adequate servers a classical application for discrete event simulation (DEVS). Therefore the frame of the model is based on Enterprise Dynamics. In order to incorporate the user-behavior the students need to be modeled which is done by an agent-based (AB) extension of the DEVS model. Reference is made at this point to Tauböck (2010), where the model is described in greater detail. The model is further extended by an (external) sub-model that simulates student walking times between courses. For this sub-model cellular automata (CA) have been used, again with an AB-extension to respect individual behavior. A thorough description of the sub-model and the coupling between the models can be found in Bruckner (2009). 3. SIMULATION REQUIREMENTS The preconditions and requirements of a simulation model imply a certain structure that has to be provided by the surrounding system (including business processes, databases, response to results, etc.). For a successful deployment of the simulation model an analysis of both, the model s needs and the peripheral system s structure, is necessary. After the individual analyses a comparison is used to reveal if a deployment is possible, or if not, what and where the obstacles are. 3.1 Structural preconditions The structural preconditions are strongly depending upon the operational mode. For example does the simulation of room shut down sequences not require any interaction from the lecturers or deans. It can be controlled by the facility management (FM) department. Since a major reconstruction process needs several months if not years of advance planning, it will be unlikely that the simulation can rely on course information and student numbers of the respective time period. Thus approximations, based on historic data will have to be used as model input. When used for day-to-day operations, on the other hand, the model will need up-to-date information from the lecturers as well as from the FM-department, requiring interfaces to connect the model with databases in which these informations are stored. Further preconditions regarding the booking procedure itself have to be met. If the peripheral system has a first come-first served booking policy installed, the potential improvement through simulation is more than limited, as the model needs a pool of requests which can be processed in order to increase efficiency. 2 As user demands will often be soft demands (e.g. get most prestigious room for lectures), the focus has to lie upon demands that can be quantified and are impartial (e.g. room capacity, infrastructure, etc.)
Fig. 1. Business process model for day-to-day use of simulation within the booking system 3.2 Business Process Modeling (BPM) Business process modeling, and especially the business process modelling notation (BPMN), prove to be helpful tools when it comes to compare model preconditions and the structure of the peripherals system. This is because of the fact, that BPMN has an appealing graphical notation [and its] basic flow elements [... ] are extremely easy to understand and grasp. This allows these models to be read and understand also by people who are not very familiar with the details of the notation (dscribed in Wolter and Schaad (2007)). 3 The major benefit of BPM is that it becomes possible to compare the necessary structure to the actual business processes of the peripheral system and that the BPM identifies all stakeholders and the interactions among each other. Especially the latter information is of benefit for acquisition of (input) data and for post-processing of simulation results to the needs of the intended audience. Figures 1 and 2 show the required structure of the peripheral system for operational deployment (fig. 1) and for 3 As the BPMN showed to bear a higher degree of complexity than needed for the task at hand, a light version with a reduced ruleset has been developed. All BPM within this paper are held in the notation of this reduced rule-set. consulting (fig. 2). As expected, when compared, they exhibit a strong influence of the operational mode (necessary depth of integration) upon the models requirements. While the business process depicted in the first model has four active stakeholders (Tutors/Lecturers, Deans, information system/tiss and the MoreSpace simulation environment) and only one passive (the FM-department which only provides information stored in its database) the situation is almost reversed in the second model. Here all actions are triggered by the FM-department and the other stakeholders only play along. This shows why it is much easier to integrate the model in the consulting mode, as it only needs input from a single database which is not controlled by the user/operator. 3.3 Data Preconditions Entity Relationship Models (ERM) Besides the structural preconditions sufficient data of adequate quality is needed for the model to work. To ensure this an entity relationship model (ERM) of the simulation model s (input and output) data structure can be set up. This procedure has the desirable side effect that this ERM can immediately be used to communicate those needs to the stakeholders in charge of the surrounding systems database(s). A valuable benefit, as it is more than
Fig. 2. Business process model for consulting deployment prior to reconstruction likely that interfaces for data import and export will have to be set up. The entity relationship diagram (ERD) in figure 3 depicts the internal ERM of the MoreSpace-tool. As can be seen, the model holds two more entities than listed in section Structure of simulation model the entities group of course and date of course. These two are merely extensions of the entity course, as courses can be split into groups and each group has at least one date on which the course takes place. 3.4 System Integration With the above methods it becomes possible to prepare the integration of the simulation model within the system. It is necessary to check... whether the preconditions of the model are met, the business process of the peripheral system is adequate, and if the necessary (input) data can be provided by the stakeholders (identified in the BPM). Good simulation models follow the credo as simple as possible, but not any simpler 4. Given a good model one may thus assume, that its requirements (preconditions, input-data, peripheral system, etc.) cannot be reduced any further. Therefore, if the requirements cannot be fulfilled, the model cannot be deployed. In such a case two alternative strategies can be pursued: (1) Reformulation of the question towards the model in order to moderate the preconditions. (2) Adaption of the business processes within the peripheral system. Since the first option leads to a model not capable of answering the original questions, it will often be undesirable. If on the other hand the costs of changing the business processes are too high, a compromise between the two might become interesting. As previously noticed, the depth of integration is connected to the levels of detail and requirements, which hints that changing the mode of operation might open the option for model deployment. Naturally this change reduces the quality and/or amount of information derived from the model. The converse argument holds as well: different simulation goals can be achieved through differing levels of system integration. In table 1 a deployment matrix is set up classifying the potential for model deployment with respect to the 4 Albert Einstein
Fig. 3. Entity relationship diagram of models internal data structure Table 1. Deployment matrix: Which deployment mode is compatible with what kind of peripheral system Operational Strategic Consulting preconditions met possible possible possible some met not possible eventually likely w/o transformation possible w/workaround none met not possible not possible not possible, w/o transformatioformatiomation w/o trans- transfor- not feasible operational mode and the condition of the peripheral system. Another very important issue of modeling, also influenced by the mode of operation, is the topic of quality management which has not been mentioned yet, although it is incorporated within the depicted BPM in figure 1 and 2 5. As an integral part of modeling and simulation (see Wenzel et al. (2008) for more details) quality management often accounts for a significant fraction of the total work load. 5 Action Does result comply with preset goals (simulationenvironment) and Evaluation of results (FM-dept.) respectively. Due to the fact that the stakeholders and components of the surrounding system can be expected to have a sound quality management installed, the necessary effort for quality management will greatly depend on the operational mode. Or, to be more precise, on the depth of system integration, as increasing depth goes hand in hand with more existing criteria to be met. 4. RESULTS & POSTPROCESSING It is not only necessary to deploy a model successfully, but also to utilize the knowledge generated by its deployment. Hence the dissemination of results can pose another obstacle. At this post-simulation stage it is crucial that generated results reach the intended stakeholders, processed in a way to extract the maximum information. Assuming the model being integrated seamlessly within the peripheral system, the first objective reaching the intended stakeholders can be taken for granted. The reason is evident when looking at the BPM in figure 1 and 2. The models describe exactly what happens with the results generated by the simulation. Both processes also include the quality management action which decides whether the output is significant or further simulation is required.
Fig. 4. Average room utilization (during fall/winter semester 2010/11) by categories of lecture room capacity. Extracting the maximum of information from the results may prove more difficult. A thorough understanding of the stakeholders needs and of the simulation outcome is essential for this task. This becomes visible when interpreting the simulation run of figure 4. While the overall room utilization of this assessment 6 lies at a low 31%, the graphic shows that the utilization is by far not equally distributed among the capacity-categories. On the contrary, there are several categories 7 which present potential bottlenecks and should probably be taken care of. 5. CONCLUSION & OUTLOOK In the course of the deployment process of the MoreSpace tool several methods for the analysis of the peripheral system have been evaluated and proved to be suited for the tasks faced. For example was it possible to derive the necessary BPM for model-deployment based on the preconditions and model requirements. Further the simulation deployment was successful in the areas where all or most requirements were met in the latter case workarounds had to be found. Nevertheless it was not yet possible to deploy the simulation model in operational (day-to-day) mode, as the adaption or transformation of business processes is a tedious process 8. Currently the post-processing of simulation results has to be done mostly manually and thus the stakeholders who receive these results are limited to the perspective of the 6 The result is obtained by real input data of the fall/winter semester 2010/2011 at TU Vienna. 7 Capacities between 60 and 100 as well as above 200 people. 8 So far only preparatory talks regarding the necessity of transformations are being held at TU Vienna. persons involved in the processing. Following the information seeking mantra overview first, zoom and filter, then details-on-demand of Shneiderman (1996), an interactive application, which enables stakeholders to actively explore the results would pose a great step forward. The implementation of an web-based application which allows for such interaction is being planned. REFERENCES Bruckner, M. (2009). Modellierung der Fußgängerdynamik im universitären Betrieb mit Hilfe von Zellulären Automaten in der Programmiersprache JAVA. Master s thesis, Vienna University of Technology, Vienna, Austria. Shneiderman, B. (1996). The eyes have it: A task by data type taxonomy for information visualizations. In P. van Zee, M. Burnett, and M. Chesire (eds.), Proceedings of the 1996 IEEE Symposium on Visual Languages, IEEE Computer Society Press, 336 343. IEEE. Tauböck, S.M. (2010). Integration of Agent Based Modelling in DEVS for Utilisation Analysis: The MoreSpace Project at TU Vienna. TU Vienna, Vienna. Wenzel, S., Weiß, M., Collisi-Böhmer, S., Pitsch, H., and Rose, O. (2008). Qualitätskriterien für die Simulation in Produktion und Logistik Planung und Durchführung von Simulationsstudien. Springer-Verlag, Berlin Heidelberg. Wiegand, D., Mebes, P., and Pichler, V. (2007). Event based simulations: Enabling improved development planning and partnership. In M. Schrenk, V.V. Popovich, and J. Benedikt (eds.), REAL CORP 007, 17 23. CORP, Wien/Schwechat, Austria. Wolter, C. and Schaad, A. (2007). Modelling of task-based authorization constraints in BPMN-based authorization constraints in BPMN. In Proceedings of 5th Intl. Conference, 64 79. Springer, Berlin Heidelberg, Deutschland.