A RAN-Level Network Slicing Dataset with SLA-Aware Feasibility Semantics

Abstract

This record releases the canonical Tier-2 Full dataset (Batch C, 100 ms aggregation window) for 5G Radio Access Network (RAN) network slicing. The dataset is generated using ns-3 and the 5G-LENA NR module and is designed to expose slice-level performance under shared and time-varying radio resources. The dataset captures scheduler outcomes and queueing dynamics through window-aggregated key performance indicators (KPIs), including observed throughput, delay, jitter, and packet loss. Measurements are reported at the slice level (URLLC, eMBB, mMTC) and reflect what the RAN actually delivers over each aggregation window rather than idealized offered rates. Each observation is associated with a binary admission-feasibility label obtained by applying fixed, slice-specific service-level agreement (SLA) checks. These labels encode whether SLA requirements are satisfied under the realized network conditions. Margin variables are included for auditing purposes but are not intended for use in machine-learning features due to potential information leakage. The dataset is constructed over a frozen scenario grid that introduces controlled variability along four dimensions: mobility (static, pedestrian, vehicular), channel conditions (LOS, LOS with shadowing, NLOS with shadowing), traffic profiles (CBR, ON/OFF), and load evolution (constant and deterministic ramp). Multiple random seeds are used to ensure statistical diversity while maintaining reproducibility. The release focuses on single-cell operation to isolate intra-cell scheduling and queueing effects. It is intended for validation-oriented research on RAN-level feasibility analysis, admission behavior, and short-horizon inference. The dataset is not designed for scheduler optimization, long-horizon forecasting, or multi-cell generalization. Only the Tier-2 Full dataset (Batch C) is released and supported for reuse. Earlier tiers (Tier-0, Tier-1, Tier-2 Pilot) and alternative window-size batches were generated exclusively for methodological validation and are not included in this release. Simulation and batch-generation scripts used to produce the dataset are provided as reproducibility artifacts. These scripts are not presented as a maintained software framework and are not required for standard dataset usage.