Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...


From Reactive to Proactive: AI-Enhanced Performance Insights in Kruize

Summary of Idea:

In dynamic containerized environments, predicting and preventing performance bottlenecks is crucial for maintaining efficiency and cost-effectiveness. Kruize provides container right-sizing recommendations based on performance metrics such as CPU and memory usage, collected from Prometheus. Currently, recommendations are generated using a percentile-based approach.

To enhance this, we propose integrating a time series and regression-based models into Kruize. These models will analyze historical performance data to generate more proactive, data-driven recommendations. The integration of these models will improve predictive accuracy, enable real-time and batch data processing, and ultimately help users optimize their infrastructure more effectively.


Project Features:

  • Collect historical usage data from Prometheus, perform feature engineering (e.g., smoothing, normalization, trend extraction) to enhance model accuracy.
  • Store processed data efficiently for real-time and batch analysis.
  • Implement and train a time series model (e.g., ARIMA, Prophet, or LSTM).
  • Develop and evaluate a regression-based model (e.g., Random Forest, XGBoost).
  • Compare their performance in terms of prediction accuracy, latency, and robustness.
  • Integrate both models into Kruize to generate performance recommendations. 


Knowledge Prerequisite:

  • Languages: Python, Java, Shell Scripting 
  • Technologies: Scikit-Learn, Pandas, NumPy, Matplotlib
  • Machine Learning Concepts: Time Series Analysis, Regression Models, Forecasting
  • Experience with containerized environments, shell scripting and performance tuning is a plus.

GitHub Repository: https://github.com/kruize/autotune.git 

Project Size: Medium (~200 Hours)

Skill Level: Intermediate

Contact(s) / Potential Mentor(s): 

Associated Community Project(s):