Benchmarks

Four-loop comparison on Linux

Measured on October 6, 2026 at commit 43f624a on an Intel Core i9-9900K, Linux 7.0.0-38-generic (x86_64), with rsloop 0.1.57 built in release mode, uvloop 0.22.1, and zuvloop 0.0.17. The GIL-enabled runs use CPython 3.14.7; the free-threaded run uses CPython 3.14.0 with the GIL disabled for all four loops. Each entry is the median of seven measured runs after two warmups, with each run in a fresh subprocess. Cyclic garbage collection is disabled during each run.

The workloads schedule 1,000,000 callbacks as one batch, run 200,000 tiny tasks that each await asyncio.sleep(0) in batches of 5,000, and perform 20,000 TCP round trips with 1,024-byte payloads. TCP uses rsloop's native fast streams; the other loops use stdlib asyncio streams.

Times are milliseconds (lower is better). Memory tables report the median of each process's peak RSS in MiB, including the interpreter and libraries. CPU affinity applies to each process and its threads. Allowing two physical cores still runs one event loop per process; it does not partition callbacks or tasks across two loops.

One core, GIL enabled (CPU 2)

Median time (ms):

Workload	asyncio	uvloop	zuvloop	rsloop
1,000,000 callbacks	550.3	258.7	179.5	182.1
200,000 tasks	566.2	341.7	307.6	325.2
20,000 TCP round trips	565.7	485.4	379.6	335.4

Peak RSS (MiB):

Workload	asyncio	uvloop	zuvloop	rsloop
1,000,000 callbacks	201.2	234.7	126.0	111.7
200,000 tasks	31.6	34.5	33.5	34.8
20,000 TCP round trips	29.8	29.8	29.8	29.8

Two cores, GIL enabled (CPUs 2–3)

Median time (ms):

Workload	asyncio	uvloop	zuvloop	rsloop
1,000,000 callbacks	674.3	260.7	184.2	195.2
200,000 tasks	571.6	350.3	305.7	324.0
20,000 TCP round trips	569.3	473.8	381.2	327.9

Peak RSS (MiB):

Workload	asyncio	uvloop	zuvloop	rsloop
1,000,000 callbacks	201.2	234.8	126.0	111.6
200,000 tasks	31.6	34.4	33.5	34.6
20,000 TCP round trips	29.7	29.7	29.7	29.7

One core, free-threaded (CPU 2)

Median time (ms):

Workload	asyncio	uvloop	zuvloop	rsloop
1,000,000 callbacks	913.1	575.9	216.5	226.8
200,000 tasks	671.1	422.3	378.7	398.3
20,000 TCP round trips	627.8	519.5	406.3	345.4

Peak RSS (MiB):

Workload	asyncio	uvloop	zuvloop	rsloop
1,000,000 callbacks	207.8	256.2	132.8	193.5
200,000 tasks	38.8	41.3	42.4	42.4
20,000 TCP round trips	36.4	36.4	36.4	36.4

Interpretation and reproduction

Zuvloop led callbacks and tasks in all three runs; rsloop led TCP. On one core with the GIL enabled, rsloop took 1.5% longer for callbacks and 5.7% longer for tasks than zuvloop, while delivering 13.2% higher TCP throughput. Its callback peak RSS was 111.7 MiB versus zuvloop's 126.0 MiB. With the GIL disabled, rsloop took 4.8% longer for callbacks and 5.2% longer for tasks, with 17.6% higher TCP throughput; its callback peak RSS rose to 193.5 MiB versus zuvloop's 132.8 MiB.

These local microbenchmarks ran on a shared host. Callback timings were particularly variable: the two-core rsloop runs ranged from 185.4 to 310.0 ms, and free-threaded asyncio ranged from 690.4 to 1,080.4 ms. The separate runs and different Python patch versions do not isolate the effects of core count, the GIL, or the latest commit. These are not general application performance claims.

To reproduce the GIL-enabled setup from the repository root:

uv venv --python 3.14.7 target/bench-gil
uv pip install --python target/bench-gil/bin/python maturin uvloop==0.22.1 zuvloop==0.0.17
VIRTUAL_ENV="$PWD/target/bench-gil" target/bench-gil/bin/maturin develop --release --locked
taskset -c 2 target/bench-gil/bin/python benches/compare_event_loops.py \
  --loops asyncio,uvloop,zuvloop,rsloop --repeat 7 --warmups 2 \
  --callbacks 1000000 --tasks 200000 --task-batch-size 5000 \
  --tcp-roundtrips 20000 --payload-size 1024 \
  --json-output target/event-loops-43f624a-one-core.json

For two physical cores on this host, replace taskset -c 2 with taskset -c 2,3 and use a different output filename. Check CPU topology before choosing affinity on another machine.

For the free-threaded setup:

uv venv --python 3.14.0t target/bench-free-threaded
uv pip install --python target/bench-free-threaded/bin/python maturin uvloop==0.22.1 zuvloop==0.0.17
VIRTUAL_ENV="$PWD/target/bench-free-threaded" target/bench-free-threaded/bin/maturin develop --release --locked
PYTHON_GIL=0 taskset -c 2 target/bench-free-threaded/bin/python benches/compare_event_loops.py \
  --loops asyncio,uvloop,zuvloop,rsloop --repeat 7 --warmups 2 \
  --callbacks 1000000 --tasks 200000 --task-batch-size 5000 \
  --tcp-roundtrips 20000 --payload-size 1024 \
  --json-output target/event-loops-43f624a-free-threaded.json