Loading...
Loading...
Contributors:
Dataset URI:
10.5281/zenodo.23435604This deposit is the version 2.0.0 delta of the replication package for an evaluation study of 5G NR recovery-layer behaviour. It accompanies the manuscript Mean Offered Load Does Not Fix the Recovery Operating Point in 5G NR: Arrival-Dependent RLC–HARQ Interaction and Per-UE Transmit-Window Saturation. The base materials — the 720-run formal campaign, its statistical outputs, the study configuration, the implementation patches and the earlier single-seed diagnostic — remain in the v1.0.0 file arrival-rlc-harq-v1.zip on this same record and are not duplicated here. v1.0.0 is unchanged. What v2.0.0 adds v2.0.0 publishes the opportunity-level attribution campaign that supports Sections 4.6–4.8 of the manuscript. Forty formal runs were replayed with their recorded commands, seeds and configuration, and every downlink scheduling opportunity was classified by the branch it took: C15 — RLC AM, HARQ off, constant-bit-rate arrivals, HIGH load — all 30 formal seeds (R11401–R11430); C14 — same configuration at MID load — 9 seeds (11301, 11306, 11311, 11312, 11320, 11322, 11325, 11327, 11328); C16 — deterministic ON–OFF arrivals at LOW load — 1 seed (11509). Across these 40 replays, 35,969,455 scheduling opportunities were observed within the measurement window, of which 8,437,422 were empty. All 8,437,422 — and none of them from any other branch — were refused by the acknowledged-mode RLC transmit window, with the fresh head at modular distance exactly 2048 from the acknowledgment lower edge in every single case. No empty opportunity was attributable to retransmission no-fit, fresh-data no-fit, STATUS no-fit, or an empty queue. Because the HIGH-load constant-bit-rate cell has exactly 30 formal runs, instrumenting all 30 constitutes a complete enumeration of that cell's prespecified runs rather than a sample of it. The 30 RngRun identities are a prespecified cohort, so the result supports statements about those 30 trajectories and not a distributional claim over unenumerated seeds. C18 R11701, which appears in Fig. 7 of the manuscript as a contrast, is the earlier independent single-seed diagnostic and is not part of this set. Contents The v2.0.0 archive is published as 86 concatenated pieces (.zip.001 – .zip.086, 12 MiB each) because a single 1 GiB transfer did not complete reliably over the authors' link. Reassemble and verify before use: cat arrival-rlc-harq-v2-gate1.zip.0* > arrival-rlc-harq-v2-gate1.zip sha256sum -c arrival-rlc-harq-v2-gate1.zip.sha256 The reassembled archive is 1,072,492,295 bytes with SHA-256 c48c3c0d4b6032f3f7ff141cb8c61a5bd81e5043fcb1e655cb3edf8cc2ec8cd1, the value recorded in arrival-rlc-harq-v2-gate1.zip.sha256. Inside the archive: runs/ — 40 run directories, each with summary.json (branch attribution, 19 counters, per-UE timelines, boundary events), opportunity-level.csv.gz (50 columns, per-opportunity records), run-provenance.json (exact command, environment, timestamps, item-by-item comparison of 63 metrics against the formal run), stdout-stderr.log, status-events.csv and gate1_record.json. audit/C15-R11401/ — one unpruned run that retains the raw mac-opportunities.csv (216 MB), phy-events.csv (89 MB), RxPacketTrace.txt, DlDataSinr.txt and rlc-backlog.csv. The production pipeline deletes the first two after analysis; this copy exists so that the analyzer's join can be re-executed by a third party. code/ — the batch driver, the branch-attribution analyzer (an out-of-tree copy of the authors' analyzer with two path substitutions), the single aggregation entry point, and the two plotting scripts used for the published figures. GATE1-SUMMARY.json — machine-readable totals per cell and per run, for readers who do not want to run any script. gate1_per_seed.csv — the per-seed table from which every per-seed number in the manuscript is computed. SHA256SUMS.txt — per-file checksums, relative paths, verifiable with sha256sum -c SHA256SUMS.txt. README.md, DATA_DICTIONARY.md, CHANGELOG-v2.0.0.md. Verification performed Every replay reproduced its formal run's 63 recorded metrics item by item; a mismatch aborts the run before attribution. The unpruned audit run was re-joined from its raw event files: the resulting summary.json is identical to the published one in a recursive field-by-field comparison, with zero differences. The 40 summary.json files aggregate to exactly the totals reported in the manuscript. The per-file checksums in SHA256SUMS.txt verify. Relation to v1.0.0 and scope v2.0.0 adds the mechanism-level attribution described above; it does not replace or modify v1.0.0, which continues to hold the 720-run metrics, the archived bootstrap contrasts, the study configuration and the implementation patches. Data and documentation are released under CC BY 4.0; software patches retain GPL-2.0-only licensing and upstream attribution. The evaluated stack is a single topology, user-equipment count, radio configuration, scheduler and application load range. A standard-conforming 18-bit sequence-number configuration was not evaluated, and no window-width sensitivity is claimed. At intermediate load nine of the thirty runs were instrumented, selected to span the range from zero to the largest observed deficit rather than drawn at random, so no occurrence frequency is estimated for that load. These limitations are stated in Section 5.7 of the manuscript.
Keyword(s):
5G NR; 5G-LENA; arrival temporal structure; HARQ; instrumented replay.
Grants and Funding:
No Grants or Funding Data Available
Associated Publication(s):
No Associated Publication Data Available
Supplementary Data:
No Supplementary Data Available
Related Dataset(s):
No Related Dataset Data Available
The National Library of Medicine (NLM) provides information about biomedical datasets. Inclusion on an NLM Web page does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: NLM Web Policies (nih.gov)