PreparedMol zstd timings
These timings measure optional zstd compression for serialized
PreparedMol payloads. Each benchmark compresses the same 1024 prepared
molecules at zstd levels 1 through 19, once without a dictionary and once with
a shipped PreparedMol dictionary.
Use these plots to choose a compression level for bulk storage. Lower compression ratio is smaller output; lower time is faster. The benchmark is indicative for this sample and machine, not a universal storage study.
The current API default for prepared.to_bytes(compression="zstd") is
dictionary_level=3, level=3. The plots below show the tradeoff space around
that default.
Benchmark setup
| Field | Value |
|---|---|
| Recorded | 2026-06-06 UTC |
| Runtime | Python 3.12.13, RDKit 2026.03.1, zstandard 0.25.0, zstd 1.5.7 |
| Platform | Linux-6.17.0-23-generic-x86_64-with-glibc2.36 |
| CPU | AMD Ryzen 5 7640U w/ Radeon 760M Graphics; 12 logical CPUs visible |
| Memory limit | 2 GiB |
| Container | compose/timings-prepared-mol-zstd.yml timings-prepared-mol-zstd service, network disabled |
| Sample | 1024 molecules, random-parseable-preparable-v1, seed 20260531 |
| Raw payload bytes | 5,742,205 |
| Training level | Dictionary artifact | Dictionary ID | Raw data |
|---|---|---|---|
| 3 | 20260531_40762836 |
1421770218 |
docs/timings-prepared-mol-zstd.tsv |
| 10 | 20260531_ebdfcd5d |
1390038199 |
docs/timings-prepared-mol-zstd-20260531_ebdfcd5d.tsv |
The plotted points are zstd levels. Error bars show standard deviation across the timing repeats recorded in the TSV.
Training level 3 dictionary
The smallest observed dictionary output in this run was about 4.8% of raw bytes at compression level 18. At compression level 10 it was about 6.1%.
Training level 10 dictionary
The smallest observed dictionary output in this run was about 4.6% of raw bytes at compression level 18. At compression level 10 it was about 5.6%.
