← Parlo
Native FUSE-T benchmark
Remote media range-cache results
Parlo mounts remote files through a native filesystem and caches the byte ranges tools touch. Cold reads include cloud fetch latency. Warm reads are served from local disk cache.
Mount
FUSE-T
fuse-t:/Parlo
File
53.9 MB MP4
real media file
Generated
2026-05-21
native mount, no WebDAV
Benchmark table
Mounted file: ~/parlo/bench-s3.mp4. Local comparison: ~/Documents/s3.mp4.
| Test | Time | Throughput | Latency |
|---|---|---|---|
| Local sequential | 0.010s | 5296.9 MiB/s | — |
| Local random seeks | 0.0007s | 9653.0 MiB/s | p50 0.0001s, p95 0.0002s |
| Local ffprobe | 0.148s | — | — |
| Parlo cold sequential | 12.55s | 4.3 MiB/s | — |
| Parlo cold random seeks | 7.95s | 0.9 MiB/s | p50 1.32s, p95 2.31s |
| Parlo cold ffprobe | 0.075s | — | — |
| Parlo warm sequential | 0.0087s | 6214.0 MiB/s | — |
| Parlo warm random seeks | 0.0011s | 6341.1 MiB/s | p50 0.0001s, p95 0.0002s |
| Parlo warm ffprobe | 0.150s | — | — |
What this proves
- • Parlo is not doing full-file sync for reads.
- • Repeated media seeks are served from local cache.
- • Warm-cache performance is effectively local for this workload.
- • The benchmark used native FUSE-T, not WebDAV.
What we are improving next
- • Faster cold reads through optimized/direct range fetch.
- • Media-aware chunk sizes and seek-neighborhood prefetch.
- • First/end-of-file pinning for MP4/MOV metadata.
- • Cache hit/miss stats in CLI and Mac dashboard.
Honest read
The architecture is right: native filesystem, byte-range reads, and local disk cache. The current cold path is still too slow for best-in-class remote editing, so the next work is cold-path optimization, larger-file tests, and media-specific prefetch policy.
CLI source command: parlo bench media ~/parlo/bench-s3.mp4 --compare-local ~/Documents/s3.mp4 --clear-cache --markdown