PeakLife (ONNX) — VM Utilization + Remaining Lifetime Predictor

This repository contains PeakLife, a lightweight neural model exported to ONNX for portable inference.
Given the historic utilization information for a VM, PeakLife predicts:

• Future CPU utilization: AvgCPU and MaxCPU (normalized)
• Remaining lifetime: normalized remaining lifetime (and seconds via scaling)

The repo includes a minimal demo pipeline that loads a small CSV (demodata.csv), preprocesses it to the model’s expected inputs, runs ONNX inference, and prints the results.

.
├── model/
│   ├── peaklife.onnx            # ONNX model
│   └── model_config.json        # Model card & Configuration
├── src/ (Optional)              # Helpers for data preparation
│   ├── DataUtil.py        
│   └── prepare_demodata.py     
├── demo.py                      # Main entry point for inference demo
├── demodata.csv                 # Demo dataset
├── requirements.txt             # Python dependencies for inference
└── README.md

This ONNX model is tied to a specific input contract and normalization:

Pre-set history length: input_length = 288 time steps by default.

• Forecast horizon: forecast_length = H.

• Signals: CPU utilization-only (AvgCPU, MaxCPU). Other resources (RAM/disk/net) are not modeled in this version.

• Normalization:

  • CPU values are expected in 0–100 in CSV and normalized by cpu_divisor (usually 100.0).

  • Remaining lifetime is normalized by max_lifetime_seconds from model_config.json.

• Output ranges: the model outputs are bounded to [0, 1] (Sigmoid heads), so it will not produce values outside this range.

• Data schema requirement for demo: demodata.csv must contain at least:

  • VMID

  • AvgCPU, MaxCPU

  • time_relative_seconds, lifetime_seconds

  • optionally TimeStamp and MaxCPU_so_far (if missing, demo falls back to last MaxCPU)

It is recommended to use a virtual environment to keep dependencies isolated.

git clone https://github.com/tgasla/MLSysOps-VM-Management-Agent.git
cd MLSysOps-VM-Management-Agent

Linux / macOS:

python3 -m venv venv
source venv/bin/activate

Windows (PowerShell):

python -m venv venv
.\venv\Scripts\Activate.ps1

pip install -r requirements.txt

Run the demo script.

python demo.py

Output Example:

--- Model Loaded ---
--- Demo Dataset Ready ---
--- PeakLife Demo ---
VMID: QdbZeJFmsJ3euIQ4lwW63NwFEP+QIirT4QbI0jEGr4dpkOet8p3iQSHAEm1gKWnR
inputs shape: (1, 288, 2) | aux shape: (1, 2)
pred_util shape: (1, 1, 2) | pred_life shape: (1, 1)

--- Utilization Prediction (AvgCPU, MaxCPU) ---
Pred (normalized): 3.912 , 5.861 | 0.039123 , 0.058608
True (normalized): 4.356 , 6.186 | 0.043559 , 0.061863

--- MAPE (avg, max, and combined) ---
util_mape(avg)=0.101832 | util_mape(max)=0.052609 | util_mape(combined)=0.077221

--- Remaining Lifetime Prediction ---
Pred remaining_lifetime_norm=0.893175 | Pred remaining_lifetime_seconds=1544747.0s
True remaining_lifetime_norm=0.950043 | True remaining_lifetime_seconds=1643100.0s

--- Lifetime MAPE ---
life_mape=0.059858

The file model/model_config.json serves as the Model Card and includes:

+ input_length

+ forecast_length

+ normalization.cpu_divisor

+ normalization.max_lifetime_seconds

+ input/output names and shapes (if you record them)

Important: The ONNX weights are tied to these dimensions and normalization constants.

If you use this repository or the provided ONNX model in your research, please cite it.

Citation metadata is available via the Zenodo DOI: 10.5281/zenodo.18422584 and in the Citation.cff file.

This repository was developed by University of Thessaly (UTH) as part of the MLSysOps project.

The MLSysOps consortium consists of 12 European partners across academia and industry, working together to optimize AI-driven operations in the Cloud-Edge Continuum.

This project has received funding from the European Union’s Horizon Europe research and innovation programme under grant agreement No 101092912.

Learn more at mlsysops.eu.