Agent-Based Simulation: Why Netlogo is running much faster than Java based Repast
Everyone says Jave is a language for large system and engineering project, which runs much faster than most other languages. I just compared it with another Agent-based modelling lanugage - Netlogo and I found Netlog runs FOUR TIMES FASTER than Jave based Repast in the classic wolf-sheep simulation model. Both models are using the same parameters for simulation and run for 5 seconds in real time. Netlogo can simulate more than 8000 time steps whereas Jave Repast can only execute for around 2600 time steps. Why?
If you are comparing Netlogo vs Repast Simphony wolf-sheep predation demos with the default settings, it is not a fair comparison because the Repast model is doing a lot more. The Repast model is performing file-based data logging, chart rendering, 2D display rendering and 3D display rendering. Both the Repast and Netlogo displays have update settings that determine how fast they are rendered relative to the tick count, and the display rendering speed is highly dependent on the GPU.
To get a better comparison of performance, we need to create a more accurate test environment. I ran the Netlogo demo using default parameters with "view updates" unchecked so the display will not update, but the chart will still update. Running the model for 20,000 ticks takes about 14 seconds. I modified the Repast demo by removing the file logging and closing the 2D and 3D displays and only leaving the chart showing, and ran the model for 20,000 ticks which also takes about 14 seconds. So the performance is exactly the same between Repast and Netlogo for this demo.
We should also consider that the demo models in both Repast and Netlogo with default parameters are TOY models with limited complexity. Typically in a more complex model that would be used in scientific studies, the agent behaviors are so complex that the individual behavior computation time is an order of magnitude greater than the toolkit framework code time, making these types of comparisons less informative of the toolkit's capability.
- Creating a bottleneck obstacle for agents
- Looking for a logic to keep a fraction in a range
- NetLogo monitor widget display changes when nothing is happening
- Netlogo: efficient iteration using 'foreach'
- NetLogo, weighted random draw from a list: how to use rnd-extension?
- Is there a way to separate the interface description from the code in net logo?
- All turtles are mating in NetLogo even after setting a condition
- NetLogo turtles leaving a trail that fades with time
- How to get a "reverse exponential distribution"?
- Using the distance variable - Robotic lawn mower simulation project
- How to assign a polygon value to existent turtles inside it with NetLogo GIS extension?
- BehaviorSpace having trouble to deal with subexperiments?
- how to define the value of a variable in an interface slider instead of agent initialization?
- How can I plot several distributions in the same window?
- LevelSpace in Netlogo: "The tick counter has not been started yet" error, but tick has indeed been reset
- I need the age of my agents before they die to create a list and histogram
- Why do I get an error when loading my .csv data into Netlogo?
- execute monthly task at end of month with netlogo time:schedule-repeating-event-with-period
- Netlogo plot shows wrong ticks
- How to record model execution time for each simulation in NetLogo?
- Netlogo divide world into upper and lower
- Extracting values from lists in Netlogo
- Netlogo BehaviorSpace reporters
- How do I use the random-seed command in Netlogo to ensure runs of my model have the reproducible results?
- How can we implement a bidimensional collision on a NetLogo Billiards?
- how does 'netlogo' 's 'any? turtles-on patch-ahead 1' command work?
- How do I export NetLogo 6.2 View when I have multiple turtle profiles?
- How to choose only one of the item to run from a list?
- NETLOGO - Agents draw random value from .csv
- Determine the Lucky PP (possible position) for each exit Netlgo evacuation simulation