Business process simulation is an important means for quantitative analysis of a business process. With the Business Process Model and Notation (BPMN) being the state-of-the-art language for the graphical representation of business processes, a variety of process simulators exist which already support the simulation of BPMN process diagrams. However, they do not provide well-defined interfaces to integrate new concepts into the simulation environment. This work demonstrates an open and extensible BPMN process simulator written in Java with defined entry points for extensions based on a plug-in structure. The demo presents the architecture of the simulator and presents how it can be extended. It is aimed at researchers in the business process management field who want to evaluate new modeling artifacts by simulation.
Business process simulation (BPS) as a quantitative analysis form for business processes
is a cost-effective way to get insights into the throughput times, costs, resource utilization
etc. of different business process designs [
BPS is also used by researchers to evaluate new modeling artifacts. However,
commercial BPMN simulators of tool vendors and also academic products, such as BIMP [
In this demo, we present an open and extensible BPMN process simulator which
builds up on a discrete simulation software called Desmo-J1. It provides a plug-in
structure with different entry points into the simulator which covers the complete simulation
flow from input parsing to output logging. The current version of the simulator consists
1 http://desmoj.sourceforge.net/
a_Architecture
of several plug-ins for advanced BPMN concept. Further, plug-ins for a new BPMN
concept, batch activities [
In the remainder, we present the basic architecture and functionality of the simulator in Section 2. Then, the plug-in structure is presented in Section 3 in which we will also describe existing plug-ins and how a new plug-in can be added. In Section 4, we draw a short conclusion. 2
The current architecture of the extensible BPMN process simulator, which is written in
Java, is shown in Fig. 1 as FMC block diagram [
Simulation User ConfGigloubraatlions Process Models
Simulation Configurations Simulation Input
R
Simulation Controller R Simulation Manager R
R R R
Log Files Output Logger
Configurations & Process Model Parser
Simulation Model Translator & Instantiator Discrete Event Simulator
R R R
Developer
Plug-ins Business Process Simulator
The Process Analyst accesses the BPMN Process Simulator via the Simulation Controller – the current user interface of the simulator shown in Fig. 2. Here, a user of the simulator can select the simulation input and can configure a simulation run. As the simulator allows simulation of multiple process models at the same time, the simulation input can consists of an arbitrary number of BPMN Process Diagrams and a Simulation Configuration for each process diagram, in which the arrival rate distribution, the activity duration distribution, the branching probability etc. is given (cf. Fig. 2). As resources can be involved in multiple processes of their organization, the resources are defined in one Global Configuration File, where the type of resources, the number of resources, and their time tables are described; concrete resources can be defined, too. To ease extensibility, all configuration files are provided in XML.
The selected simulation input is loaded by the Simulation Manager which accesses all other components of the process simulator. First, it calls the Configurations & Process Model Parser consisting of several parsers to convert the different simulation input files into internal structures which can be efficiently queried during the simulation. With this, the Simulation Model Translator & Instantiator is called which is responsible to translate the simulation input into a DES simulation model and to initialize the DES experiment executing the simulation model. Therefore, first, the parsed input is translated into static model components like queues, distributions and structures for data collection. Second, the initial state of the simulation model is instantiated. Initial events, e.g., an event representing the generation of process instance, for the event list of the Discrete Event Simulator, are created and activated.
From the insights collected by the Simulation Manager, the Output Logger creates the artificial process logs and a simulation report including several calculated key performance indicators (KPIs), such as the average throughput time. Regarding the extensibility, well-defined entry points are desired for the architecture of the process simulator. We provide the concept of Plug-ins, allowing the development of new functionality for the process simulator. A plug-in refers to one single feature and can be switched on and off in the Simulation Controller (cf. Fig. 2) before simulation, thus supporting modularity on simulation usage level. By switching a plug-in on, its code is additionally executed at the defined entry point. The plug-in concept will be presented in next section.
Details on the actual prototypical implementation of the extensible BPMN process
simulator called Scylla2 can be found in [
The demonstrated process simulator is open-source, but provides also a list of abstractions which can be refined to extend the process simulator with new concepts.
Therefore, several abstract classes are defined as entry points for developers to write plug-ins. In Fig. 3, they are categorized into the different stages of simulation: parsing, initialization, execution, and reporting. The ParserPluggable offers an entry point to the input parsers, such that new simulation input, e.g., a new BPMN element, can be parsed. During initialization of the DES simulation model, the DESMO-J distribution for the arrival rate and activity distributions are set. The entry point DistributionConversionPluggable allows the initialization of additional distributions. While executing a 2 Its source code is available at https://github.com/bptlab/scylla and a screen cast is available at https://bpt.hpi.uni-potsdam.de/Public/scylla.
simulation experiment, events are generated, stored in queues, and if an event occurs, their event routines are executed which usually results in new events. Different entry points are available to influence DES events: EventCreationPluggable to generate new type of events, EventSchedulingPluggable to influence the scheduling of events, and EventOrderTypePluggable to adapt the priorities of events and changing their order in the queues. For influencing the implemented BPMN behavior of the simulator, two entry points exists: one on the process instance level – ProcessInstanceGenerationEventPluggable– and one on the BPMN events level – BPMNConstructsPluggable. The latter ones includes several sub-classes to influence the behavior of the minimum set of BPMN elements supported by the basic simulator, for instance, BPMNStartEventPluggable, TaskEnableEventPluggable. The ExternalEventPluggable offers the opportunity to add behavior which is not strictly related to a single process instance, but to the general behavior of business process simulation.
Fig. 4 shows which
BPMN concepts are sup- 0% 25% 50% 75%
ported by the core process Normal FTlaoswk
simulator (shown in bold), End Event
whereas constructs sup- (None) Start Event
ported as plug-ins are not Data-BasedPXoOoRl
emphasized. These were Start Message
selected based on their TeMxteAsnsangoetaFtlioown
usage frequency by prac- Parallel Fork/Join
titioners [
For extending the simulator, a plug-in developer can use the presented pluggableclasses to extend them for modifying the simulator at different stages. Therefore, a developer would generate a package to which all plug-in classes and additional needed classes are added. This package can be added as jar-file to the simulator-project. The new plug-in package, then, has to be registered in the plug-in list nMETA-INFnpluginsnplugins list which can be found in the project resources. 4
This paper presented an extensible BPMN process simulator with its architecture and its plug-in structure to extend and modify the basic BPMN simulator. The current simulator already includes plug-in packages for several advanced BPMN constructs which can be selected or deselected for a simulation run. Further, the entry points to the simulator were used to integrate a new BPMN concept – batch activities for batch processing in business processes. In future, we aim to extend the simulator user interface to show the generated simulator reports graphically.
A special thanks to the student Leon Bein which developed the simulator user interface.