Michel Gokan Khan Javid Taheri Auday Al-Dulaimy Andreas Kassler
Karlstad University, Karlstad, Sweden
| 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.
Note
The performance modeling part required to create the service chain models is not included in this suite. Users will need to prepare their models based on their own data or existing benchmarks.
Caution
The current open source version of this package is a mere research prototype and is not intended for production use. It is provided as-is, without any guarantees or warranties. Please do not consider this package as a fully functional product at the moment. We are actively working on improving the package and adding new features, but we cannot provide any support for it at this time. There may or may not be a commercial version available soon, but we cannot guarantee. Use at your own risk. If you have any questions or concerns, please open an issue on the GitHub repository.
Caution
This package contains several bugs and issues that we are actively working on fixing. Please be aware that the package may not work as expected in all cases. We are working on resolving these issues and will provide updates as soon as possible. If you encounter any problems, please open an issue on the GitHub repository, so we know.
- 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}}
- Clone PerfSim repository by running the following command:
git clone [email protected]:michelgokan/perfsim.git
-
Navigate to the PerfSim directory:
cd perfsim -
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:
-
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
testsdirectory, where you can run as follows:pytest test_single_thread.py::Test1SFC1S1R1T1HBE::test_all_traffic_types_all_topologies -
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.pyThis 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
-
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.shPerfSim requires the following to run:
- Python 3.12 or later
- Additional Python libraries as listed in the
requirements.txtfile.
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!
