SDRBench — XGC

Gyrokinetic fusion plasma simulation (XGC, PPPL): fields on an unstructured mesh of 20694 nodes x 512 toroidal planes, 9 time steps, double precision.

Raw binary arrays, not an Arrow/Parquet dataset. datasets.load_dataset("sdrbench/xgc") returns the table of files (files.jsonl: path, dtype, shape, sha256). Read the arrays with the sdrbench package or hf_hub_download + numpy as shown below.

This repository is an unmodified mirror of the XGC data of SDRBench, the Scientific Data Reduction Benchmark. The originals are hosted by Argonne National Laboratory on Globus; every archive was unpacked and its files uploaded byte-for-byte (sha256 verified) (revision f2518596a1). metadata/ holds the original SDRBench property and template files. Kept in sync automatically by szcompressor/sdrbench.

• Data provider: Princeton Plasma Physics Laboratory (PPPL); https://jychoi-hpc.github.io/adios-python-docs/XGC-mesh-and-field-data.html
• Size: 3.1 GB in 2 variant(s)
• Format: raw little-endian binary, C order (slowest dimension first); dtype and shape per field below

Usage

pip install "sdrbench[sz3]"     # sdrbench + pysz (SZ3)

import numpy as np
import sdrbench
from pysz import sz, szConfig, szErrorBoundMode

ds = sdrbench.dataset("xgc")                # default variant "raw"
print(ds.fields)
x = ds["mesh_coordinates-values"]                     # numpy array, dtype <f8, shape (20694, 2)

# compress with SZ3 (pysz) at a 1e-3 value-range-relative error bound
conf = szConfig()
conf.errorBoundMode = szErrorBoundMode.REL
conf.relErrorBound = 1e-3
compressed, ratio = sz.compress(np.ascontiguousarray(x), conf)
y, _ = sz.decompress(compressed, x.dtype.type, x.shape)
max_err, psnr, nrmse = sz.verify(np.asarray(x), y)
print(f"ratio {ratio:.1f}x, PSNR {psnr:.1f} dB, max error {max_err:.3g}")

Files are fetched on first use (in parallel with ds.download("data/")) and cached, pinned to the revision of the installed sdrbench release; if Hugging Face is unreachable the package falls back to the original archive on Globus. Existing local SDRBench copies can be used with sdrbench.dataset("xgc", root="/path").

Without the package, with huggingface_hub and numpy only:

from huggingface_hub import hf_hub_download
import numpy as np
p = hf_hub_download("sdrbench/xgc", "xgc/xgc.mesh.bp_coordinates-values_20694x2_double", repo_type="dataset", revision="f2518596a116a47c2e439417e17f4d9458012ccb")
x = np.fromfile(p, dtype="<f8").reshape((20694, 2))

Variants and fields

raw (default)

sdrbench.dataset("xgc", "raw") — folder xgc/ — 45 files, 1.5 GB, from xgc.zip.

Raw binary arrays extracted from the ADIOS files: per time step dpot, iden (20694 x 512), pot0, potm0 (20694); mesh arrays prefixed mesh_.

13 variable(s) x 9 steps (08100 ... 08900): ds.series("dpot"), ds["dpot", "08100"].

Field	dtype	Shape (C order)	File
dpot/08100	<f8	20694 x 512	xgc/xgc.3d.08100.bp_dpot_20694x512_double
iden/08100	<f8	20694 x 512	xgc/xgc.3d.08100.bp_iden_20694x512_double
pot0/08100	<f8	20694	xgc/xgc.3d.08100.bp_pot0_20694_double
potm0/08100	<f8	20694	xgc/xgc.3d.08100.bp_potm0_20694_double
dpot/08200	<f8	20694 x 512	xgc/xgc.3d.08200.bp_dpot_20694x512_double
iden/08200	<f8	20694 x 512	xgc/xgc.3d.08200.bp_iden_20694x512_double
pot0/08200	<f8	20694	xgc/xgc.3d.08200.bp_pot0_20694_double
potm0/08200	<f8	20694	xgc/xgc.3d.08200.bp_potm0_20694_double
dpot/08300	<f8	20694 x 512	xgc/xgc.3d.08300.bp_dpot_20694x512_double
iden/08300	<f8	20694 x 512	xgc/xgc.3d.08300.bp_iden_20694x512_double
... 35 more

adios

sdrbench.dataset("xgc", "adios") — folder xgc.3d/ — 10 files, 1.5 GB, from xgc.3d.tar.gz.

The original ADIOS BP files (not raw arrays; read them with ADIOS).

Other files: xgc.3d/xgc.3d.08100.bp, xgc.3d/xgc.3d.08200.bp, xgc.3d/xgc.3d.08300.bp, xgc.3d/xgc.3d.08400.bp, xgc.3d/xgc.3d.08500.bp, xgc.3d/xgc.3d.08600.bp, xgc.3d/xgc.3d.08700.bp, xgc.3d/xgc.3d.08800.bp, xgc.3d/xgc.3d.08900.bp, xgc.3d/xgc.mesh.bp

Citation

Please cite SDRBench; sdrbench cite xgc prints it:

@inproceedings{zhao2020sdrbench,
  title     = {{SDRBench}: Scientific Data Reduction Benchmark for Lossy Compressors},
  author    = {Zhao, Kai and Di, Sheng and Liang, Xin and Li, Sihuan and Tao, Dingwen and Chen, Zizhong and Cappello, Franck},
  booktitle = {2020 IEEE International Conference on Big Data (Big Data)},
  pages     = {2716--2724},
  year      = {2020}
}

License and terms

The data are distributed by the SDRBench team for research use under the terms of the original data providers listed above (see the SDRBench website); please acknowledge them as requested. Rights remain with the data providers.