Benchmarks¶

Suite at benchmarks/.

Run¶

# terminal 1
cortex-discovery

# terminal 2
python benchmarks/bench_all.py --output results.json

Individual benchmarks:

• bench_latency.py — one-way publisher→subscriber latency (async).
• bench_latency_sync.py — raw sync zmq baseline (no asyncio).
• bench_latency_inproc.py — in-process pub + sync sub + a CPU-bound asyncio neighbour, to expose GIL contention.
• bench_throughput.py — messages/sec and MB/sec.
• bench_all.py — full matrix; JSON output via --output.

Reading results¶

• p99 is what matters for real-time workloads; the mean hides jitter.
• For array workloads, MB/s approaching memcpy bandwidth means zero-copy transport is working.
• Serialization overhead via inproc sockets with copy=False is reported separately to isolate encode/decode from the network path.