SDRBench — Miranda

Rayleigh-Taylor instability simulated with LLNL's Miranda hydrodynamics code.

Raw binary arrays, not an Arrow/Parquet dataset. datasets.load_dataset("sdrbench/miranda") 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 Miranda 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 cd543db67f). metadata/ holds the original SDRBench property and template files. Kept in sync automatically by szcompressor/sdrbench.

• Data provider: LLNL; contact Peter Lindstrom (lindstrom2@llnl.gov)
• Size: 118.1 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("miranda")                # default variant "small"
print(ds.fields)
x = ds["density"]                     # numpy array, dtype <f8, shape (256, 384, 384)

# 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("miranda", 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/miranda", "256x384x384/density.d64", repo_type="dataset", revision="cd543db67fcb47de3c645312f2db9a27198a9af7")
x = np.fromfile(p, dtype="<f8").reshape((256, 384, 384))

Variants and fields

small (default)

sdrbench.dataset("miranda", "small") — folder 256x384x384/ — 7 files, 2.1 GB, from SDRBENCH-Miranda-256x384x384.tar.gz.

7 fields of a 256 x 384 x 384 subdomain at an early time, double precision. pressure, viscocity and diffusivity are legacy (incorrect) names of time-derivative restart fields.

Field	dtype	Shape (C order)	File
density	<f8	256 x 384 x 384	256x384x384/density.d64
diffusivity	<f8	256 x 384 x 384	256x384x384/diffusivity.d64
pressure	<f8	256 x 384 x 384	256x384x384/pressure.d64
velocityx	<f8	256 x 384 x 384	256x384x384/velocityx.d64
velocityy	<f8	256 x 384 x 384	256x384x384/velocityy.d64
velocityz	<f8	256 x 384 x 384	256x384x384/velocityz.d64
viscocity	<f8	256 x 384 x 384	256x384x384/viscocity.d64

big

sdrbench.dataset("miranda", "big") — folder 3072x3072x3072/ — 98 files, 116.0 GB, from SDRBENCH-Miranda-3072x3072x3072.tar.gz.

One 3072^3 single-precision density field at a late time, stored as 96 slabs density/00 ... density/95 of 32 x 3072 x 3072; ds.series('density').concatenate() joins them (116 GB).

1 variable(s) x 96 steps (00 ... 95): ds.series("density"), ds["density", "00"].

Field	dtype	Shape (C order)	File
density/00	<f4	32 x 3072 x 3072	3072x3072x3072/density-00.f32
density/01	<f4	32 x 3072 x 3072	3072x3072x3072/density-01.f32
density/02	<f4	32 x 3072 x 3072	3072x3072x3072/density-02.f32
density/03	<f4	32 x 3072 x 3072	3072x3072x3072/density-03.f32
density/04	<f4	32 x 3072 x 3072	3072x3072x3072/density-04.f32
density/05	<f4	32 x 3072 x 3072	3072x3072x3072/density-05.f32
density/06	<f4	32 x 3072 x 3072	3072x3072x3072/density-06.f32
density/07	<f4	32 x 3072 x 3072	3072x3072x3072/density-07.f32
density/08	<f4	32 x 3072 x 3072	3072x3072x3072/density-08.f32
density/09	<f4	32 x 3072 x 3072	3072x3072x3072/density-09.f32
... 86 more

Other files: 3072x3072x3072/SDRBENCH-Miranda-3072x3072x3072-property.txt, 3072x3072x3072/template_data-3072x3072x3072.txt

Citation

Please cite SDRBench; sdrbench cite miranda 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.