| Webpage | arXiv | Dataset | Presentation video (my PhD defence) | PerfSim in IEEE Xplore

PerfSim is a discrete-event simulator designed to approximate and predict the performance of cloud-native service chains in various user-defined scenarios. It leverages a systematic approach for analyzing network traces and simulating the behavior of service chains to provide insights into system KPIs, such as the average response time of requests.

• Discrete-Event Simulation: Offers precise simulation of cloud-native microservice behaviors over time.
• User-Defined Scenarios: Allows users to simulate custom scenarios, including different service chain configurations and resource allocation policies.
• High Simulation Accuracy: Achieves 81-99% simulation accuracy in predicting the average latency of incoming requests compared to real Kubernetes deployments.
• Resource Efficiency: Designed to run on modest hardware, such as a single laptop, facilitating large-scale simulations without the need for a real testbed.

While PerfSim provides valuable insights into the performance of cloud-native microservice chains, it has the following limitations:

• Modeling Exclusion: PerfSim does not include a performance modeling component. Users must supply their own microservice performance models in JSON format or in the code.
• Simulation Complexity: For highly complex service chains with intricate dependencies and interactions, simulation results may vary from real-world deployments due to the abstraction level.
• Resource Intensive Scenarios: Extremely resource-intensive simulations may require hardware beyond a single laptop for optimal performance.
• Networking Details: Simplified networking simulation may not capture all nuances of real-world network behavior, particularly under specific or extreme conditions.

If you use PerfSim in your scientific work, please consider citing our original paper:

M. Gokan Khan, J. Taheri, A. Al-Dulaimy and A. Kassler, "PerfSim: A Performance Simulator for Cloud Native Microservice Chains," in IEEE Transactions on Cloud Computing, vol. 11, no. 2, pp. 1395-1413, 1 April-June 2023, doi: 10.1109/TCC.2021.3135757

Bibtex:

@ARTICLE{9652084,
  author={Gokan Khan, Michel and Taheri, Javid and Al-Dulaimy, Auday and Kassler, Andreas},
  journal={IEEE Transactions on Cloud Computing}, 
  title={PerfSim: A Performance Simulator for Cloud Native Microservice Chains}, 
  year={2023},
  volume={11},
  number={2},
  pages={1395-1413},
  keywords={Cloud computing;Computational modeling;Microservice architectures;Resource management;Emulation;Containers;Testing;Performance simulator;performance modeling;cloud native computing;service chains;simulation platform},
  doi={10.1109/TCC.2021.3135757}}

1. Clone PerfSim repository by running the following command:

git clone git@github.com:michelgokan/perfsim.git

2. Navigate to the PerfSim directory:

   cd perfsim
   

3. Install dependencies (ensure you have Python 3.8 or later installed):

   pip install -r requirements.txt
   

At the moment, there are 3 ways to run PerfSim scenarios:

1. Run PerfSim via writing a Python script: Users can write a Python script to define the simulation scenarios and run the simulation. There are examples under tests directory, where you can run as follows:

   pytest test_single_thread.py::Test1SFC1S1R1T1HBE::test_all_traffic_types_all_topologies
   

2. Run PerfSim via web API(Flask): Users can run PerfSim via a web API. To run the web API, execute the following command:

   python server.py
   

   This will start the PerfSim server, and you can access via the following endpoints:

   • /perfsim/: To check if the server is running
   • /perfsim/api/v1/config/setupAll: To setup the simulation
   • /perfsim/api/v1/scenario/run: To run the simulation
   • /perfsim/api/v1/scenario/saveAll: To save the simulation results

3. Run PerfSim via command line: Users can run PerfSim via the command line. To run PerfSim via the command line, execute the following command:

   python perfsim.py --config <config_file_path> --scenario-id <scenario_id>
   

   e.g.,:

   python perfsim.py --config-path examples/example.json --scenario-id 1
   

You can then analyze the simulation results generated by PerfSim for performance insights.

To rebuild the documentation, make sure to first install sphinx and then run the following command:

cd docs
./rebuild_docs.sh

PerfSim requires the following to run:

• Python 3.12 or later
• Additional Python libraries as listed in the requirements.txt file.

We welcome contributions to PerfSim! If you have suggestions for improvements or bug fixes, please fork the repository and submit a pull request.

PerfSim is released under the GPL V2 License. See the LICENSE file for more details.

• Michel Gokan Khan (main contributor)
• Maybe yourself? Send your first pull request to be listed here!