SNDZoo dataset used for VNF workload forecasting

This dataset was presented in the following paper submitted (under revision) to the SAYGreeN 2026 workshop, co-located with IEEE PerCom 2026 conference: C. Puliafito, A. Virdis, C. Vallati, N. Tonellotto, M. Galante, G. Anastasi, E. Mingozzi, M. Cucurachi, and L. Martorana, “Using Foundation Models to Forecast the Workload of Edge Network Functions for Green Computing”.

This is the dataset generated through the SNDZoo tng-bench framework (https://github.com/sonata-nfv/tng-sdk-benchmark) and used for evaluating performance of Lag-Llama and TimesFM time-series foundation models toward forecasting of VNF workload in cloud-edge environments.

The dataset comprises data about three distinct VNFs:
– Nginx (WEB)
– Mosquitto (IOT)
– Snort (SEC)
and four different metrics:
– CPU
– memory
– received data
– transmitted data

To replicate the experiments carried out to generate this dataset, please follow the tutorial at https://github.com/sonata-nfv/tng-sdk-benchmark/wiki/Setup-execution-platform-(vim-emu).

This repository contains two directories:
– RAW_SNDZoo_Experiments, which includes the dataset raw data, prior to interpolation and normalization
– DatasetSNDZoo, which includes the final preprocessed data, which are then given as input to Lag-Llama and TimesFM models

Each of the two directories contains a README file further detailing each of them.