Skip to content

Performance Benchmarks

Jump to bottom
whisprer edited this page Oct 5, 2025 · 4 revisions

Performance Benchmarks

Methodology

  • Command: secure_delete <file> --passes N --pattern random|zeros

  • Timer: shell time (real/user/sys)

  • File sizes: 1 GiB for single-pass zeros; 1 GiB × 3 passes for random test (3 GiB total writes)

  • Host: ThinkPad P52, Intel i7 @ 4.2 GHz, 64 GB RAM, dual NVMe

Generate test data





# 1 GiB random file
dd if=/dev/urandom of=testfile.bin bs=1M count=1024 status=progress

Run (examples)





# 3 random passes
time secure_delete testfile.bin --passes 3 --pattern random
# 1 zero pass
dd if=/dev/urandom of=testfile.bin bs=1M count=1024 status=none
time secure_delete testfile.bin --passes 1 --pattern zeros

Log results to benchmarks/results.md for longitudinal tracking.

Native vs WSL/NTFS Results

(ThinkPad P52, i7 @ 4.2 GHz, 64 GB RAM, dual NVMe)

Filesystem / Path | Command | Passes | Pattern | real | user | sys | Throughput -- | -- | -- | -- | -- | -- | -- | -- WSL → NTFS (/mnt/d) | secure_delete testfile.bin | 3 | random | 24.84s | 0.11s | 7.60s | ≈ 130 MB/s WSL ext4 (~/bench) | same | 3 | random | 10.24s | 0.13s | 8.15s | ≈ 315 MB/s ext4 (zeros pattern) | --passes 1 --pattern zeros | 1 | zeros | 1.07s | 0.00s | 0.55s | ≈ 960 MB/s dd direct write | dd oflag=direct | — | — | 1.39s | — | — | ≈ 3.1 GB/s dd direct read | dd iflag=direct | — | — | 0.79s | — | — | ≈ 5.5 GB/s

Interpretation

  • Native ext4 recovers near-NVMe throughput; WSL → NTFS adds 2–3× latency due to boundary flushes.

  • Per-pass sync_all() adds roughly 5–10 % latency — expected and by design.

  • Random × 3 passes writes 3× the bytes; apparent MB/s drops vs single-pass zeros, trading speed for entropy.

  • Bottom line: the Rust shred core is I/O-bound, not CPU-bound. Keep targets on native Linux/ext4 when possible.

Clone this wiki locally