|Author||David E. Sorokin|
|Keywords||simulation discrete event simulation simulation library queue network|
Using Aivika Modeler, you can create quite fast discrete event simulation models that are translated into native code. Also you can run the simulation experiments by the Monte Carlo method, specifying that how the results should be processed. It can plot Time Series, Deviation chart by the confidence interval, plot histograms, save the results in the CSV files for the further analysis and more. All is defined in just a few lines of code written in Python. Then the report of the simulation experiment with charts, statistics summary and links to the saved CSV files is automatically opened in your Web browser.
To take a taste of Aivika Modeler, here is a complete simulation model and the corresponding experiment that define a simple queue network. The model contains a transact generator, two bounded queues, two servers and the arrival timer that measures the processing of transacts. The experiment launches 1000 simulation runs in parallel, plots charts and then opens a report with the results of simulation in the Web browser. The compilation, simulation and chart plotting took about 1 minute on my laptop.
Example: Work Stations in Series
This is a model of two work stations connected in a series and separated by finite queues. It is described in different sources [1, 2]. So, this is chapter 7 of  and section 5.14 of .
 A. Alan B. Pritsker, Simulation with Visual SLAM and AweSim, 2nd ed.
 Труб И.И., Объектно-ориентированное моделирование на C++: Учебный курс. - СПб.: Питер, 2006
The maintenance facility of a large manufacturer performs two operations. These operations must be performed in series; operation 2 always follows operation 1. The units that are maintained are bulky, and space is available for only eight units including the units being worked on. A proposed design leaves space for two units between the work stations, and space for four units before work station 1. [..] Current company policy is to subcontract the maintenance of a unit if it cannot gain access to the in-house facility.
Historical data indicates that the time interval between requests for maintenance is exponentially distributed with a mean of 0.4 time units. Service times are also exponentially distributed with the first station requiring on the average 0.25 time units and the second station, 0.5 time units. Units are transported automatically from work station 1 to work station 2 in a negligible amount of time. If the queue of work station 2 is full, that is, if there are two units awaiting for work station 2, the first station is blocked and a unit cannot leave the station. A blocked work station cannot server other units.
#!/usr/local/bin/python3 from simulation.aivika.modeler import * model = MainModel() # the transacts can have assignable and updatable fields, but it is not used here data_type = TransactType(model, 'Transact') # it will help us to measure the processing time of transacts timer = create_arrival_timer(model, name = 'timer', descr = 'Measures the processing time') timer_source = timer.add_result_source() # this is a generator of transacts input_stream = exponential_random_stream(data_type, 0.4) # a queue before the first workstation queue1 = create_queue(model, data_type, 4, name = 'queue1', descr = 'Queue no. 1') queue1_source = queue1.add_result_source() # another queue before the second workstation queue2 = create_queue(model, data_type, 2, name = 'queue2', descr = 'Queue no. 2') queue2_source = queue2.add_result_source() # the first workstation activity is modeled by the server workstation1 = exponential_random_server(data_type, 0.25, name = 'workstation1', descr = 'Workstation no. 1') workstation1_source = workstation1.add_result_source() # this is the second workstation workstation2 = exponential_random_server(data_type, 0.5, name = 'workstation2', descr = 'Workstation no. 2') workstation2_source = workstation2.add_result_source() # try to enqueue the arrivals; otherwise, count them as lost enqueue_stream_or_remove_item(queue1, input_stream) # a chain of streams originated from the first queue stream2 = dequeue_stream(queue1) stream3 = server_stream(workstation1, stream2) enqueue_stream(queue2, stream3) # another chain of streams, which must be terminated already stream4 = dequeue_stream(queue2) stream5 = server_stream(workstation2, stream4) stream5 = arrival_timer_stream(timer, stream5) terminate_stream(stream5) # reset the statistics after 30 time units reset_time = 30 reset_queue(queue1, reset_time) reset_queue(queue2, reset_time) reset_server(workstation1, reset_time) reset_server(workstation2, reset_time) reset_arrival_timer(timer, reset_time) # it defines the simulation specs specs = Specs(0, 300, 0.1) processing_factors = [workstation1_source.processing_factor, workstation2_source.processing_factor] # define what to display in the report views = [ExperimentSpecsView(), InfoView(), FinalStatsView(title = 'Processing Time (Statistics Summary)', series = [timer_source.processing_time]), DeviationChartView(title = 'Processing Factor (Chart)', right_y_series = processing_factors), FinalHistogramView(title = 'Processing Factor (Histogram)', series = processing_factors), FinalStatsView(title = 'Processing Factor (Statistics Summary)', series = processing_factors), FinalStatsView(title = 'Lost Items (Statistics Summary)', series = [queue1_source.enqueue_lost_count]), DeviationChartView(title = 'Queue Size (Chart)', right_y_series = [queue1_source.count, queue2_source.count]), FinalStatsView(title = 'Queue Size (Statistics Summary)', series = [queue1_source.count_stats, queue2_source.count_stats]), DeviationChartView(title = 'Queue Wait Time (Chart)', right_y_series = [queue1_source.wait_time, queue2_source.wait_time]), FinalStatsView(title = 'Queue Wait Time (Statistics Summary)', series = [queue1_source.wait_time, queue2_source.wait_time])] # it will render the report renderer = ExperimentRendererUsingDiagrams(views) # it defines the simulation experiment with 1000 runs experiment = Experiment(renderer, run_count = 1000) # it compiles the model and runs the simulation experiment model.run(specs, experiment)
After running the simulation experiment, you will see the Deviation charts that will show the confidence intervals by rule 3 sigma. Also you will see a general information about the experiment as well as histograms and summary statistics sections for some properties such as the queue size, queue wait time, the processing time of transacts and the server processing factor in the final time point.
The model written in Python is translated into its Haskell representation based on using the Aivika simulation libraries, namely aivika and aivika-transformers. Then the translated model is compiled by GHC into native code and executed. The simulation itself should be quite fast and efficient.
For the first time, the process of compiling and preparing the model for running may take a few minutes. On next time, it may take just a few seconds.
There is one prerequisite, though. To use Aivika Modeler, you must have Stack installed on your computer. The main operating systems are supported: Windows, Linux and macOS.
Then you can install the aivika-modeler package using pip in usual way.
Aivika Modeler is licensed under the open-source BSD3 license like that how the main libraries of Aivika itself are licensed under this license.
In most cases you do not need to know the Haskell programming language. The knowledge of Python will be sufficient to create and run many simulation models. But if you will need a non-standard component, for example, to simulate the TCP/IP protocol, then you or somebody else will have to write its implementation in Haskell and then create the corresponding wrapper in Python so that it would be possible to use the component from Python.
There is a separation of concerns. Python is used as a high-level glue for combining components to build the complete simulation model, while Haskell is used as a high-level modeling language for writing such components.
Aivika itself also supports a DSL, which is very similar to the popular GPSS modeling language but not fully equivalent, though. This DSL is implemented in package aivika-gpss. There are plans to add the corresponding support to Aivika Modeler too. Please stay tuned.
You can find a more full information on website www.aivikasoft.com.