Fibre Cache ZIPF benchmarks
Zipf-distributed reads over ten times the capacity, inserting on a miss
Throughput in Mops/s: Operations completed per second across all workers over a 10 s run at capacity 10000. Measured on MacBook M4 Pro (14 cores).
The best result in each row is highlighted, leaving out quick_cache, which is a yardstick and is not ranked. Italics mean quick_cache is above the best ranked engine. A dash means the engine has no API for that mode.
The mixed 64 row is from a rerun on 2026-10-31 at load average 6.87, after the full run gave fibre_cache.s3fifo 16.66 and fibre_cache.tinylfu 11.35 and neither repeated.
sync
| Workers | fibre_cache.s3fifo | fibre_cache.tinylfu | moka | quick_cache | stretto |
|---|---|---|---|---|---|
| 8 | 30.15 | 23.57 | 5.57 | 42.44 | 3.66 |
| 64 | 31.34 | 23.91 | 1.11 | 21.02 | 3.59 |
async
| Workers | fibre_cache.s3fifo | fibre_cache.tinylfu | moka | quick_cache | stretto |
|---|---|---|---|---|---|
| 8 | 30.72 | 20.88 | 5.39 | 47.01 | 3.86 |
| 64 | 31.28 | 25.87 | 4.01 | 24.85 | 3.75 |
mixed
| Workers | fibre_cache.s3fifo | fibre_cache.tinylfu | quick_cache |
|---|---|---|---|
| 64 | 26.20 | 19.39 | 21.30 |
Hit ratio and peak entries
| Mode | Workers | fibre_cache.s3fifo | fibre_cache.tinylfu | moka | quick_cache | stretto |
|---|---|---|---|---|---|---|
| sync | 8 | 0.777 / 9984 | 0.774 / 9984 | 0.788 / 10000 | 0.785 / 10048 | 0.652 / 10002 |
| sync | 64 | 0.778 / 9984 | 0.774 / 9984 | 0.785 / 10000 | 0.785 / 10048 | 0.615 / 10003 |
| async | 8 | 0.777 / 9984 | 0.774 / 9984 | 0.787 / 10000 | 0.785 / 10048 | 0.668 / 10002 |
| async | 64 | 0.777 / 9984 | 0.774 / 9984 | 0.788 / 10000 | 0.785 / 10048 | 0.650 / 10002 |
| mixed | 64 | 0.777 / 9984 | 0.774 / 9984 | - | 0.785 / 10048 | - |
Hit ratio over the run, then the highest entry count the sampler saw against the capacity.