Benchmarks

async-cache ships a benchmark suite in benchmarks/run.py that measures real-world performance characteristics.

Running Benchmarks

# Default (10,000 iterations)
python benchmarks/run.py

# Custom iterations
python benchmarks/run.py --iterations 50000

# Also write JSON results
python benchmarks/run.py --json results.json

What Is Measured

Hit vs Miss Latency — per-operation latency for cache hits and misses
Thundering Herd — 100 coroutines racing on 1 key; verifies only 1 loader fires
Batch DataLoader — throughput of batch_loader mode (keys/sec)
Competitor Comparison — side-by-side with aiocache.SimpleMemoryCache and cachetools.LRUCache (sync baseline)

Sample Results

Python 3.12, 10,000 iterations, Linux

Hit vs Miss Latency

Metric	Hit	Miss
Median latency	~0.8µs	~3.2µs
p99 latency	~2.8µs	~11µs
Throughput	>1M ops/s	>300K ops/s

Thundering Herd Protection

Metric	Value
Coroutines per round	100
Loader calls per round	1.0 (verified)
Round time	~600µs

Batch DataLoader

Metric	Value
Keys per batch	200
Throughput	~60K keys/s

Competitor Comparison

Library	Avg Hit Latency	Notes
async-cache	~0.8µs	async-native, thundering herd, batching
aiocache	varies	Install `aiocache` to compare
cachetools (sync)	varies	Sync-only, no herd/batch protection

To include competitor results, install the optional packages:

pip install aiocache cachetools
python benchmarks/run.py