# zoo-bench performance report

raw data: `results/zoo-framework-0.8.0/20261011T092340Z.json`

## Run environment

| item | value |
|---|---|
| CPU model | AMD EPYC 7763 64-Core Processor |
| logical cores | 4 |
| platform | Linux-x86_64 |
| operating system | Linux 6.17.0-1022-azure (#22-Ubuntu SMP Mon Jul 27 17:24:03 UTC 2026) |
| Python | CPython 3.13.16 |
| interpreter | /opt/hostedtoolcache/Python/3.13.16/x64/bin/python |
| GIL | default |
| framework under test (distribution metadata) | 0.8.0 |
| framework under test (module __version__) | 0.8.0 (footnote only, not the version criterion) |
| drive-surface generation of the framework under test | current-generation drive surface (scheduling core + scheduling model; the completion signal comes from the result reactor) |
| harness | zoo-bench 0.0.1 @ 9a42be24ac43f75e2e5d7b16434f91ba3d0717b6 |
| command to reproduce | `zoo-bench run --framework zoo-framework==0.8.0` |

> dist_version (the distribution metadata) is the source of truth; module_version is only a footnote

## Workload

- composition: one unit of work = JSON serialize + deserialize + string concatenation, repeated until the target duration is reached; the tiers are declared by body_tiers_us in matrix.yaml
- **the workload is a stand-in, not a real trace**: its shape resembles business actions such as 'take a frame of data -> serialize -> parse -> transform strings', but this repository holds no real trace, so every conclusion is conditional on that

## Measurement methodology

- **the body duration is instrumented within the same run**: the body measures itself internally and reports back; framework overhead = end-to-end - it. **No cross-run subtraction**  -  two runs differ in state (caches, frequencies, scheduling noise), so subtracting introduces systematic bias.
- **warmups are excluded from statistics**: each unit first runs 3 warmup rounds that are discarded, then only the following 30 measured rounds are counted; the sample count equals 'measured rounds x concurrency'.
- **absolute durations are not comparable across runs**: they only show the order of magnitude; across versions/machines read the **relative quantities within the same run** (overhead share, multiples against the baselines), which are insensitive to overall speed.

## Conclusion summary

- Once the framework under test zoo runs a body of about 2700 microseconds or more, its framework overhead drops below 15% of end-to-end; below that tier, the main price of choosing it is the framework overhead itself.
- At concurrency 1 and a 40 microsecond body, the framework under test is 2.12x faster than process_pool (the fastest measured tier at that concurrency).
- At concurrency 1, once the body exceeds about 40 microseconds, apscheduler is no longer faster than the framework under test.
- At concurrency 4, once the body exceeds about 40 microseconds, apscheduler is no longer faster than the framework under test.
- At concurrency 16, once the body exceeds about 40 microseconds, apscheduler is no longer faster than the framework under test.
- At concurrency 64, once the body exceeds about 40 microseconds, apscheduler is no longer faster than the framework under test.
- At concurrency 1, once the body exceeds about 40 microseconds, asyncio_pool is no longer faster than the framework under test.
- At concurrency 4, within the measured tiers asyncio_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 16, within the measured tiers asyncio_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 64, within the measured tiers asyncio_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 1, within the measured tiers bare_thread is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 4, within the measured tiers bare_thread is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 16, within the measured tiers bare_thread is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 64, within the measured tiers bare_thread is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 1, once the body exceeds about 40 microseconds, process_pool is no longer faster than the framework under test.
- At concurrency 4, within the measured tiers process_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 16, within the measured tiers process_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 64, once the body exceeds about 40 microseconds, process_pool is no longer faster than the framework under test.
- At concurrency 1, within the measured tiers thread_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 4, within the measured tiers thread_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 16, within the measured tiers thread_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 64, within the measured tiers thread_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.

> a single-point speedup has no selection meaning; the conclusion is phrased as 'which scheme to choose at what body duration' (the same framework overhead is 70% of a 40 microsecond task and a few percent of a 10 millisecond one)

### Overhead threshold crossings

| scheme | concurrency | threshold | first tier at or below the threshold (microseconds) | measured tiers (microseconds) | note |
|---|---|---|---|---|---|
| apscheduler | 1 | 15.00% | 2700 | 40, 300, 2700, 10000 | this is the smallest measured tier that satisfies the condition; the true crossing point lies between it and the previous tier |
| apscheduler | 4 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| apscheduler | 16 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| apscheduler | 64 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| asyncio_pool | 1 | 15.00% | 2700 | 40, 300, 2700, 10000 | this is the smallest measured tier that satisfies the condition; the true crossing point lies between it and the previous tier |
| asyncio_pool | 4 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| asyncio_pool | 16 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| asyncio_pool | 64 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| bare_thread | 1 | 15.00% | 2700 | 40, 300, 2700, 10000 | this is the smallest measured tier that satisfies the condition; the true crossing point lies between it and the previous tier |
| bare_thread | 4 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| bare_thread | 16 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| bare_thread | 64 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| process_pool | 1 | 15.00% | 2700 | 40, 300, 2700, 10000 | this is the smallest measured tier that satisfies the condition; the true crossing point lies between it and the previous tier |
| process_pool | 4 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| process_pool | 16 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| process_pool | 64 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| thread_pool | 1 | 15.00% | 300 | 40, 300, 2700, 10000 | this is the smallest measured tier that satisfies the condition; the true crossing point lies between it and the previous tier |
| thread_pool | 4 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| thread_pool | 16 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| thread_pool | 64 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| zoo | 1 | 15.00% | 2700 | 40, 300, 2700, 10000 | this is the smallest measured tier that satisfies the condition; the true crossing point lies between it and the previous tier |
| zoo | 4 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| zoo | 16 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| zoo | 64 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |

### Relative turning points

| baseline | concurrency | first tier where it is no longer faster (microseconds) | note |
|---|---|---|---|
| apscheduler | 1 | 40 | this is the smallest measured tier that satisfies the condition |
| apscheduler | 4 | 40 | this is the smallest measured tier that satisfies the condition |
| apscheduler | 16 | 40 | this is the smallest measured tier that satisfies the condition |
| apscheduler | 64 | 40 | this is the smallest measured tier that satisfies the condition |
| asyncio_pool | 1 | 40 | this is the smallest measured tier that satisfies the condition |
| asyncio_pool | 4 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| asyncio_pool | 16 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| asyncio_pool | 64 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| bare_thread | 1 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| bare_thread | 4 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| bare_thread | 16 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| bare_thread | 64 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| process_pool | 1 | 40 | this is the smallest measured tier that satisfies the condition |
| process_pool | 4 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| process_pool | 16 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| process_pool | 64 | 40 | this is the smallest measured tier that satisfies the condition |
| thread_pool | 1 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| thread_pool | 4 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| thread_pool | 16 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| thread_pool | 64 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |

## Charts

![overhead_ratio](figures/en-overhead_ratio.svg)

> Body tiers at concurrency 4 only; ratios hold within a single run. The overhead numbers of concurrency 16, 64 were withheld (beyond this machine's parallelism; see the caliber section) and are not charted

![throughput](figures/en-throughput.svg)

> Body tier 40 microseconds only; throughput holds within a single run

![relative_multiple](figures/en-relative_multiple.svg)

> One panel per measured concurrency (4 in total); multiples hold within a single run

## Dimension: Latency percentiles and jitter

> end-to-end / concurrency; percentiles and relative spread are given in the same table  -  a metric that only gives the median is not acceptable

| concurrency | body tier (microseconds) | zoo | apscheduler | asyncio_pool | bare_thread | process_pool | thread_pool |
|---|---|---|---|---|---|---|---|
| 1 | 40 | 174.06 microseconds | 281.24 microseconds | 213.36 microseconds | 144.76 microseconds | 369.75 microseconds | 77.28 microseconds |
| 1 | 300 | 433.22 microseconds | 544.52 microseconds | 405.28 microseconds | 396.70 microseconds | 614.59 microseconds | 324.43 microseconds |
| 1 | 2700 | 2.800 milliseconds | 2.884 milliseconds | 2.701 milliseconds | 2.680 milliseconds | 2.982 milliseconds | 2.658 milliseconds |
| 1 | 10000 | 9.885 milliseconds | 10.156 milliseconds | 9.700 milliseconds | 9.548 milliseconds | 10.087 milliseconds | 9.698 milliseconds |
| 4 | 40 | 200.10 microseconds | 255.42 microseconds | 156.05 microseconds | 168.63 microseconds | 191.30 microseconds | 96.24 microseconds |
| 4 | 300 | 457.43 microseconds | 516.26 microseconds | 407.59 microseconds | 412.07 microseconds | 341.25 microseconds | 352.35 microseconds |
| 4 | 2700 | 2.814 milliseconds | 2.983 milliseconds | 2.764 milliseconds | 2.728 milliseconds | 1.359 milliseconds | 2.711 milliseconds |
| 4 | 10000 | 10.050 milliseconds | 10.101 milliseconds | 9.786 milliseconds | 9.829 milliseconds | 4.650 milliseconds | 9.811 milliseconds |
| 16 | 40 | 199.72 microseconds | 246.53 microseconds | 130.72 microseconds | 170.14 microseconds | 152.53 microseconds | 88.97 microseconds |
| 16 | 300 | 464.10 microseconds | 520.81 microseconds | 392.41 microseconds | 419.21 microseconds | 277.10 microseconds | 354.95 microseconds |
| 16 | 2700 | 2.869 milliseconds | 2.998 milliseconds | 2.756 milliseconds | 2.682 milliseconds | 1.413 milliseconds | 2.133 milliseconds |
| 16 | 10000 | 10.104 milliseconds | 10.223 milliseconds | 9.931 milliseconds | 9.895 milliseconds | 4.698 milliseconds | 9.734 milliseconds |
| 64 | 40 | 195.85 microseconds | 245.59 microseconds | 123.00 microseconds | 174.50 microseconds | 200.04 microseconds | 90.65 microseconds |
| 64 | 300 | 465.14 microseconds | 514.29 microseconds | 381.41 microseconds | 426.28 microseconds | 310.51 microseconds | 353.92 microseconds |
| 64 | 2700 | 2.881 milliseconds | 2.940 milliseconds | 2.804 milliseconds | 2.775 milliseconds | 1.375 milliseconds | 2.774 milliseconds |
| 64 | 10000 | 10.631 milliseconds | 9.958 milliseconds | 10.397 milliseconds | 9.824 milliseconds | 4.637 milliseconds | 10.331 milliseconds |

per-unit values (percentiles, spread, raw per-round samples) are in 'Appendix: all values' at the end.

## Dimension: Framework overhead share

> overhead = end-to-end / concurrency - the body duration measured within the same run (no cross-run subtraction). **cells with -**: that concurrency exceeds this machine's parallel capacity; the body's self-reported duration contains scheduling waits from oversubscription and is not comparable with the row's end-to-end per task, so no overhead figure is given (end-to-end and throughput live in their own dimensions)

| concurrency | body tier (microseconds) | zoo | apscheduler | asyncio_pool | bare_thread | process_pool | thread_pool |
|---|---|---|---|---|---|---|---|
| 1 | 40 | 71.17% | 82.72% | 69.72% | 63.87% | 84.78% | 39.13% |
| 1 | 300 | 30.28% | 45.98% | 26.53% | 25.54% | 49.47% | 11.12% |
| 1 | 2700 | 4.86% | 10.26% | 4.58% | 4.26% | 11.34% | 1.81% |
| 1 | 10000 | 1.50% | 3.60% | 1.58% | 1.33% | 3.39% | 0.73% |
| 4 | 40 | 72.44% | 78.18% | 65.64% | 69.06% | 53.06% | 48.39% |
| 4 | 300 | 33.04% | 40.22% | 25.72% | 28.67% | - | 15.31% |
| 4 | 2700 | 7.13% | 9.79% | 4.02% | 4.69% | - | 3.67% |
| 4 | 10000 | - | - | - | - | - | - |
| 16 | 40 | - | - | - | - | - | - |
| 16 | 300 | - | - | - | - | - | - |
| 16 | 2700 | - | - | - | - | - | - |
| 16 | 10000 | - | - | - | - | - | - |
| 64 | 40 | - | - | - | - | - | - |
| 64 | 300 | - | - | - | - | - | - |
| 64 | 2700 | - | - | - | - | - | - |
| 64 | 10000 | - | - | - | - | - | - |

per-unit values (percentiles, spread, raw per-round samples) are in 'Appendix: all values' at the end.

## Dimension: Throughput and concurrency scaling

> throughput = concurrency / median end-to-end; whether it keeps rising with concurrency is the scaling behaviour

| concurrency | body tier (microseconds) | zoo | apscheduler | asyncio_pool | bare_thread | process_pool | thread_pool |
|---|---|---|---|---|---|---|---|
| 1 | 40 | 5745.3 | 3555.7 | 4686.8 | 6907.9 | 2704.5 | 12939.1 |
| 1 | 300 | 2308.3 | 1836.5 | 2467.4 | 2520.8 | 1627.1 | 3082.3 |
| 1 | 2700 | 357.2 | 346.7 | 370.2 | 373.2 | 335.3 | 376.2 |
| 1 | 10000 | 101.2 | 98.5 | 103.1 | 104.7 | 99.1 | 103.1 |
| 4 | 40 | 19989.9 | 15660.3 | 25632.9 | 23720.3 | 20909.5 | 41562.7 |
| 4 | 300 | 8744.6 | 7748.0 | 9813.8 | 9707.1 | 11721.5 | 11352.2 |
| 4 | 2700 | 1421.5 | 1340.8 | 1447.4 | 1466.1 | 2944.3 | 1475.5 |
| 4 | 10000 | 398.0 | 396.0 | 408.8 | 407.0 | 860.2 | 407.7 |
| 16 | 40 | 80111.5 | 64899.7 | 122396.1 | 94038.7 | 104898.1 | 179841.5 |
| 16 | 300 | 34475.3 | 30721.3 | 40773.2 | 38167.2 | 57740.1 | 45076.5 |
| 16 | 2700 | 5576.0 | 5337.6 | 5805.4 | 5965.9 | 11327.3 | 7501.0 |
| 16 | 10000 | 1583.6 | 1565.0 | 1611.1 | 1617.0 | 3406.0 | 1643.7 |
| 64 | 40 | 326787.3 | 260593.3 | 520342.3 | 366766.2 | 319932.6 | 705985.9 |
| 64 | 300 | 137591.9 | 124444.1 | 167797.3 | 150135.6 | 206112.0 | 180832.2 |
| 64 | 2700 | 22213.7 | 21770.5 | 22823.4 | 23067.0 | 46544.6 | 23074.4 |
| 64 | 10000 | 6020.1 | 6427.2 | 6155.5 | 6514.5 | 13803.0 | 6194.7 |

per-unit values (percentiles, spread, raw per-round samples) are in 'Appendix: all values' at the end.

## Dimension: Overhead attribution

> scope: splits the per-task end-to-end into submit side / handoff / body / return and, at the same tier and concurrency, **compares each segment against every baseline**, answering 'which segment the excess falls on'. It is **diagnostic, not the main evidence**: it samples a small set of tiers, and does not compare total overhead against the baselines. **A negative handoff is not an error**: it means the submit call had not returned yet while the body was already running  -  schemes that count thread creation as submit side behave this way (measured: bare_thread handoff -343 microseconds). **The return segment includes waiting for the completion signal** (the last stretch before draining returns); it shares the main measurement's scope

**status: ok**

> this dimension is a **diagnosis**, not the main evidence: the sampled tiers and concurrencies are a fixed small batch, not the whole matrix. The concurrency ceiling is the machine's parallel capacity - beyond it, what is measured is contention;

sampled tiers (microseconds): 40, 10000; concurrency: [1, 4]; this machine's parallel capacity: 4

| scheme | concurrency | tier (microseconds) | submit side | handoff | body | return | status |
|---|---|---|---|---|---|---|---|
| apscheduler | 1 | 40 | 107.12 microseconds | 80.51 microseconds | 57.45 microseconds | 41.85 microseconds | ok |
| apscheduler | 4 | 40 | 89.24 microseconds | 511.93 microseconds | 58.52 microseconds | 235.26 microseconds | ok |
| apscheduler | 1 | 10000 | 180.64 microseconds | 129.18 microseconds | 9.644 milliseconds | 63.22 microseconds | ok |
| apscheduler | 4 | 10000 | 128.64 microseconds | 10.573 milliseconds | 14.891 milliseconds | 9.882 milliseconds | ok |
| asyncio_pool | 1 | 40 | 17.11 microseconds | 25.55 microseconds | 47.21 microseconds | 58.41 microseconds | ok |
| asyncio_pool | 4 | 40 | 17.90 microseconds | 290.31 microseconds | 55.58 microseconds | 275.71 microseconds | ok |
| asyncio_pool | 1 | 10000 | 22.76 microseconds | 35.65 microseconds | 9.380 milliseconds | 84.75 microseconds | ok |
| asyncio_pool | 4 | 10000 | 24.43 microseconds | 12.321 milliseconds | 14.391 milliseconds | 9.524 milliseconds | ok |
| bare_thread | 1 | 40 | 142.34 microseconds | -75.72 microseconds | 52.62 microseconds | 30.13 microseconds | ok |
| bare_thread | 4 | 40 | 149.38 microseconds | -73.41 microseconds | 50.50 microseconds | 299.38 microseconds | ok |
| bare_thread | 1 | 10000 | 5.174 milliseconds | -5094.15 microseconds | 9.410 milliseconds | 37.10 microseconds | ok |
| bare_thread | 4 | 10000 | 5.181 milliseconds | -5083.47 microseconds | 14.567 milliseconds | 9.130 milliseconds | ok |
| process_pool | 1 | 40 | - | - | - | - | not_measurable: the body's start/end timestamps are unavailable - it did not run in this process (cross-process units cannot be split this way) |
| process_pool | 4 | 40 | - | - | - | - | not_measurable: the body's start/end timestamps are unavailable - it did not run in this process (cross-process units cannot be split this way) |
| process_pool | 1 | 10000 | - | - | - | - | not_measurable: the body's start/end timestamps are unavailable - it did not run in this process (cross-process units cannot be split this way) |
| process_pool | 4 | 10000 | - | - | - | - | not_measurable: the body's start/end timestamps are unavailable - it did not run in this process (cross-process units cannot be split this way) |
| thread_pool | 1 | 40 | 8.07 microseconds | 10.50 microseconds | 44.59 microseconds | 15.18 microseconds | ok |
| thread_pool | 4 | 40 | 11.09 microseconds | 187.33 microseconds | 50.57 microseconds | 155.11 microseconds | ok |
| thread_pool | 1 | 10000 | 11.95 microseconds | 16.89 microseconds | 9.546 milliseconds | 27.48 microseconds | ok |
| thread_pool | 4 | 10000 | 13.35 microseconds | 10.158 milliseconds | 18.653 milliseconds | 9.036 milliseconds | ok |
| zoo | 1 | 40 | 51.07 microseconds | 25.19 microseconds | 52.73 microseconds | 64.23 microseconds | ok |
| zoo | 4 | 40 | 66.95 microseconds | 358.54 microseconds | 55.25 microseconds | 319.43 microseconds | ok |
| zoo | 1 | 10000 | 87.78 microseconds | 31.70 microseconds | 9.652 milliseconds | 97.67 microseconds | ok |
| zoo | 4 | 10000 | 82.24 microseconds | 10.273 milliseconds | 14.148 milliseconds | 9.899 milliseconds | ok |

### where the excess over each baseline falls

| baseline | concurrency | tier (microseconds) | excess per task | mostly on | gap in that segment |
|---|---|---|---|---|---|
| apscheduler | 1 | 40 | -99.89 microseconds | return | 22.38 microseconds |
| asyncio_pool | 1 | 40 | 42.90 microseconds | submit side | 33.97 microseconds |
| bare_thread | 1 | 40 | 43.72 microseconds | handoff | 100.91 microseconds |
| thread_pool | 1 | 40 | 114.47 microseconds | return | 49.05 microseconds |
| apscheduler | 4 | 40 | -138.41 microseconds | return | 84.17 microseconds |
| asyncio_pool | 4 | 40 | 78.10 microseconds | handoff | 68.23 microseconds |
| bare_thread | 4 | 40 | 342.78 microseconds | handoff | 431.96 microseconds |
| thread_pool | 4 | 40 | 373.11 microseconds | handoff | 171.21 microseconds |
| apscheduler | 1 | 10000 | -151.45 microseconds | return | 34.45 microseconds |
| asyncio_pool | 1 | 10000 | 336.94 microseconds | body | 271.67 microseconds |
| bare_thread | 1 | 10000 | 341.61 microseconds | handoff | 5.126 milliseconds |
| thread_pool | 1 | 10000 | 264.05 microseconds | body | 106.11 microseconds |
| apscheduler | 4 | 10000 | -764.89 microseconds | return | 16.48 microseconds |
| asyncio_pool | 4 | 10000 | 434.00 microseconds | return | 374.94 microseconds |
| bare_thread | 4 | 10000 | 9.711 milliseconds | handoff | 15.357 milliseconds |
| thread_pool | 4 | 10000 | 671.38 microseconds | return | 862.47 microseconds |

> instrumentation seams used (temporarily wrapped on the harness side, no framework code changed): BaseWaiter._dispatch_worker (handing a worker to the scheduling model); WorkerDispatchCore.resolve_period / resolve_phase / resolve_run_timeout (per-round policy lookups); BaseWaiter.execute_service (one scheduling round)

> **the cost of instrumentation itself** (on the framework-under-test side, paired from two rounds in the same subprocess, so it can be discounted honestly): zoo 40 microseconds / concurrency 1: without instrumentation 190.97 microseconds vs with 193.02 microseconds, a difference of 2.05 microseconds; zoo 40 microseconds / concurrency 4: without instrumentation 749.61 microseconds vs with 769.88 microseconds, a difference of 20.27 microseconds; zoo 10000 microseconds / concurrency 1: without instrumentation 9.779 milliseconds vs with 9.868 milliseconds, a difference of 89.14 microseconds; zoo 10000 microseconds / concurrency 4: without instrumentation 39.848 milliseconds vs with 39.597 milliseconds, a difference of 251.34 microseconds (faster with instrumentation: this round's machine drift exceeded the instrumentation cost)

> **additivity of the sub-measurements**: the sub-measurements are **independent measurements**: each one's duration contains its own preemption waits, so their sum can exceed the submit side they all live in (about 3% measured). Do not read them as an additive partition.measured overlap this round: zoo 40 microseconds / concurrency 4 overlaps 55.7%; zoo 10000 microseconds / concurrency 1 overlaps 16.8%; zoo 10000 microseconds / concurrency 4 overlaps 97.9%

### what the submit side of the framework under test consists of

- concurrency 1, tier 40 microseconds: rest of the scheduling round (due checks, locks, list upkeep) 2.16 microseconds; dispatch (handed to the scheduling model, incl. pool enqueue) 24.60 microseconds; per-round policy lookups (period / phase / timeout each read config) 17.00 microseconds; submit side, other (the adapter's own bookkeeping and completion-signal state) 7.30 microseconds
- concurrency 4, tier 40 microseconds: rest of the scheduling round (due checks, locks, list upkeep) 0.00 microseconds; dispatch (handed to the scheduling model, incl. pool enqueue) 37.56 microseconds; per-round policy lookups (period / phase / timeout each read config) 30.47 microseconds; submit side, other (the adapter's own bookkeeping and completion-signal state) 5.07 microseconds
- concurrency 1, tier 10000 microseconds: rest of the scheduling round (due checks, locks, list upkeep) 3.41 microseconds; dispatch (handed to the scheduling model, incl. pool enqueue) 41.16 microseconds; per-round policy lookups (period / phase / timeout each read config) 29.77 microseconds; submit side, other (the adapter's own bookkeeping and completion-signal state) 12.94 microseconds
- concurrency 4, tier 10000 microseconds: rest of the scheduling round (due checks, locks, list upkeep) 0.00 microseconds; dispatch (handed to the scheduling model, incl. pool enqueue) 46.12 microseconds; per-round policy lookups (period / phase / timeout each read config) 34.15 microseconds; submit side, other (the adapter's own bookkeeping and completion-signal state) 7.59 microseconds

### Attribution conclusions

- At concurrency 1 and a 40 microsecond body, the framework under test is not slower overall than apscheduler (-100 microseconds per task): the two sides have different timing structures, and the per-segment differences have both signs, so no 'which segment' is given.
- At concurrency 1 and a 40 microsecond body, the framework under test spends 43 microseconds more per task than asyncio_pool, most of it on 'submit side' (+34 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'handoff' is -0 microseconds, so read them together.
- At concurrency 1 and a 40 microsecond body, the framework under test spends 44 microseconds more per task than bare_thread, most of it on 'handoff' (+101 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'submit side' is -91 microseconds, so read them together.
- At concurrency 1 and a 40 microsecond body, the framework under test spends 114 microseconds more per task than thread_pool, most of it on 'return' (+49 microseconds).
- In this group the framework's submit side consists of (microseconds): rest of the scheduling round (due checks, locks, list upkeep) 2, dispatch (handed to the scheduling model, incl. pool enqueue) 25, per-round policy lookups (period / phase / timeout each read config) 17, submit side, other (the adapter's own bookkeeping and completion-signal state) 7.
- At concurrency 4 and a 40 microsecond body, the framework under test is not slower overall than apscheduler (-138 microseconds per task): the two sides have different timing structures, and the per-segment differences have both signs, so no 'which segment' is given.
- At concurrency 4 and a 40 microsecond body, the framework under test spends 78 microseconds more per task than asyncio_pool, most of it on 'handoff' (+68 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'body' is -0 microseconds, so read them together.
- At concurrency 4 and a 40 microsecond body, the framework under test spends 343 microseconds more per task than bare_thread, most of it on 'handoff' (+432 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'submit side' is -82 microseconds, so read them together.
- At concurrency 4 and a 40 microsecond body, the framework under test spends 373 microseconds more per task than thread_pool, most of it on 'handoff' (+171 microseconds).
- In this group the framework's submit side consists of (microseconds): rest of the scheduling round (due checks, locks, list upkeep) 0, dispatch (handed to the scheduling model, incl. pool enqueue) 38, per-round policy lookups (period / phase / timeout each read config) 30, submit side, other (the adapter's own bookkeeping and completion-signal state) 5.
- At concurrency 1 and a 10000 microsecond body, the framework under test is not slower overall than apscheduler (-151 microseconds per task): the two sides have different timing structures, and the per-segment differences have both signs, so no 'which segment' is given.
- At concurrency 1 and a 10000 microsecond body, the framework under test spends 337 microseconds more per task than asyncio_pool, most of it on 'body' (+272 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'handoff' is -4 microseconds, so read them together.
- At concurrency 1 and a 10000 microsecond body, the framework under test spends 342 microseconds more per task than bare_thread, most of it on 'handoff' (+5126 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'submit side' is -5087 microseconds, so read them together.
- At concurrency 1 and a 10000 microsecond body, the framework under test spends 264 microseconds more per task than thread_pool, most of it on 'body' (+106 microseconds).
- In this group the framework's submit side consists of (microseconds): rest of the scheduling round (due checks, locks, list upkeep) 3, dispatch (handed to the scheduling model, incl. pool enqueue) 41, per-round policy lookups (period / phase / timeout each read config) 30, submit side, other (the adapter's own bookkeeping and completion-signal state) 13.
- At concurrency 4 and a 10000 microsecond body, the framework under test is not slower overall than apscheduler (-765 microseconds per task): the two sides have different timing structures, and the per-segment differences have both signs, so no 'which segment' is given.
- At concurrency 4 and a 10000 microsecond body, the framework under test spends 434 microseconds more per task than asyncio_pool, most of it on 'return' (+375 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'handoff' is -2047 microseconds, so read them together.
- At concurrency 4 and a 10000 microsecond body, the framework under test spends 9711 microseconds more per task than bare_thread, most of it on 'handoff' (+15357 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'submit side' is -5098 microseconds, so read them together.
- At concurrency 4 and a 10000 microsecond body, the framework under test spends 671 microseconds more per task than thread_pool, most of it on 'return' (+862 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'body' is -4505 microseconds, so read them together.
- In this group the framework's submit side consists of (microseconds): rest of the scheduling round (due checks, locks, list upkeep) 0, dispatch (handed to the scheduling model, incl. pool enqueue) 46, per-round policy lookups (period / phase / timeout each read config) 34, submit side, other (the adapter's own bookkeeping and completion-signal state) 8.

## Dimension: The cost of scheduling semantics

> scope: this dimension uses an **internal switch comparison** (same workload, semantics all off vs. enabled one by one) and **does not compare across schemes**  -  a bare writeup has no counterpart for priority / timeout / retry, so it can only answer 'what is this framework's semantics worth', not 'how much more expensive than a bare writeup'

**status: ok** (the conclusion was reached on the 'current' drive-surface generation installed during this measurement)

| semantics item | layer | how probed | measurable | evidence / reason |
|---|---|---|---|---|
| timeout | dispatch layer | behavioral probe (driven repeatedly through scheduling rounds; measures the moment the worker leaves the in-flight table) | yes | timeout reap observed: the worker left the in-flight table before the body ended (0.0123s) (body 0.15s, run_timeout 0.01s); judgment action at dispatch_core.py:365; what is in effect is observation and circuit-breaking (recording, marking unhealthy, removing the in-flight registration, stopping dispatch), **not termination** of the still-running worker - CPython cannot safely interrupt a running thread |
| priority | event layer | structural probe (checks whether the symbols required for execution exist) | no | not re-checked: this item's conclusion was reached on the 'previous' generation; taking it as this generation's fact asserts wiring in the report that the current generation does not have. Re-deriving it needs a probe design of its own (each item needs clean on/off states before it can be priced) |
| retry | event layer | structural probe (checks whether the symbols required for execution exist) | no | not re-checked: this item's conclusion was reached on the 'previous' generation; taking it as this generation's fact asserts wiring in the report that the current generation does not have. Re-deriving it needs a probe design of its own (each item needs clean on/off states before it can be priced) |

> this conclusion was reached on the version under test and its coordinates hold for that version only; after appending a version to matrix.yaml, re-run this probe

## Dimension: Native task line: three execution forms of the same subject

> scope: this line is measured on its own: it enters neither the main matrix's pivot tables nor its crossings, and its numbers must not be read against the baseline adapters listed there

**verdict: not_applicable**

the framework under test has no such capability (symbols not found: adaptive_params, dual_arm_worker, native_params, native_task_worker) - this line does not apply to this version, and that is not a failure

equivalence gate: not run (the probe did not reach it)

> loaded task modbus_rtu.parse_response - Modbus RTU response-frame parsing: the input is a bare RTU ADU response frame (no transport framing), and the tier's frame bytes are its length

> tiers R1, R8, R125 (frame bytes 7, 21, 255); concurrencies 1, 4, 16, 64; each cell runs 100 warm-up plus 100 measured rounds, one child process per concurrency (4 in all)

## tiers where the framework under test is faster

> Tiers where the framework under test is ahead (the two sides differ by more than that tier's band). This comes from the **same within-run comparison** as the openly published unfavourable data  -  one side takes the faster tiers, the other the slower ones, so both cover the same ground and neither can pick its own tiers. **This section being empty is not a failure**: it should be empty when nothing is distinguishable or when the framework loses everywhere.

| baseline | concurrency | body tier (microseconds) | gap | median of the framework under test |
|---|---|---|---|---|
| apscheduler | 1 | 40 | the framework under test is 1.62x faster than apscheduler | 174.06 microseconds |
| process_pool | 1 | 40 | the framework under test is 2.12x faster than process_pool | 174.06 microseconds |
| apscheduler | 1 | 300 | the framework under test is 1.26x faster than apscheduler | 433.22 microseconds |
| process_pool | 1 | 300 | the framework under test is 1.42x faster than process_pool | 433.22 microseconds |
| apscheduler | 1 | 2700 | the framework under test is 1.03x faster than apscheduler | 2.800 milliseconds |
| process_pool | 1 | 2700 | the framework under test is 1.07x faster than process_pool | 2.800 milliseconds |
| apscheduler | 1 | 10000 | the framework under test is 1.03x faster than apscheduler | 9.885 milliseconds |
| process_pool | 1 | 10000 | the framework under test is 1.02x faster than process_pool | 9.885 milliseconds |
| apscheduler | 4 | 40 | the framework under test is 1.28x faster than apscheduler | 200.10 microseconds |
| apscheduler | 4 | 300 | the framework under test is 1.13x faster than apscheduler | 457.43 microseconds |
| apscheduler | 4 | 2700 | the framework under test is 1.06x faster than apscheduler | 2.814 milliseconds |
| apscheduler | 16 | 40 | the framework under test is 1.23x faster than apscheduler | 199.72 microseconds |
| apscheduler | 16 | 300 | the framework under test is 1.12x faster than apscheduler | 464.10 microseconds |
| apscheduler | 16 | 2700 | the framework under test is 1.04x faster than apscheduler | 2.869 milliseconds |
| apscheduler | 64 | 40 | the framework under test is 1.25x faster than apscheduler | 195.85 microseconds |
| apscheduler | 64 | 300 | the framework under test is 1.11x faster than apscheduler | 465.14 microseconds |

## Openly published unfavourable data

> Tiers where the framework under test is behind (the two sides differ by more than that tier's band). **A report missing this part is not acceptable**  -  a report that only shows where it wins gives the reader every reason to distrust it. Tiers whose difference is smaller than the band are not here but in the 'no winner' section: that is the verdict's outcome, not a place to hide things.

| baseline | concurrency | body tier (microseconds) | gap | median of the framework under test |
|---|---|---|---|---|
| thread_pool | 1 | 40 | thread_pool is 2.25x faster than zoo | 174.06 microseconds |
| thread_pool | 1 | 300 | thread_pool is 1.34x faster than zoo | 433.22 microseconds |
| asyncio_pool | 1 | 2700 | asyncio_pool is 1.04x faster than zoo | 2.800 milliseconds |
| bare_thread | 1 | 2700 | bare_thread is 1.04x faster than zoo | 2.800 milliseconds |
| thread_pool | 1 | 2700 | thread_pool is 1.05x faster than zoo | 2.800 milliseconds |
| asyncio_pool | 1 | 10000 | asyncio_pool is 1.02x faster than zoo | 9.885 milliseconds |
| bare_thread | 1 | 10000 | bare_thread is 1.04x faster than zoo | 9.885 milliseconds |
| thread_pool | 1 | 10000 | thread_pool is 1.02x faster than zoo | 9.885 milliseconds |
| asyncio_pool | 4 | 40 | asyncio_pool is 1.28x faster than zoo | 200.10 microseconds |
| thread_pool | 4 | 40 | thread_pool is 2.08x faster than zoo | 200.10 microseconds |
| asyncio_pool | 4 | 300 | asyncio_pool is 1.12x faster than zoo | 457.43 microseconds |
| bare_thread | 4 | 300 | bare_thread is 1.11x faster than zoo | 457.43 microseconds |
| process_pool | 4 | 300 | process_pool is 1.34x faster than zoo | 457.43 microseconds |
| thread_pool | 4 | 300 | thread_pool is 1.30x faster than zoo | 457.43 microseconds |
| bare_thread | 4 | 2700 | bare_thread is 1.03x faster than zoo | 2.814 milliseconds |
| process_pool | 4 | 2700 | process_pool is 2.07x faster than zoo | 2.814 milliseconds |
| thread_pool | 4 | 2700 | thread_pool is 1.04x faster than zoo | 2.814 milliseconds |
| asyncio_pool | 4 | 10000 | asyncio_pool is 1.03x faster than zoo | 10.050 milliseconds |
| bare_thread | 4 | 10000 | bare_thread is 1.02x faster than zoo | 10.050 milliseconds |
| process_pool | 4 | 10000 | process_pool is 2.16x faster than zoo | 10.050 milliseconds |
| thread_pool | 4 | 10000 | thread_pool is 1.02x faster than zoo | 10.050 milliseconds |
| asyncio_pool | 16 | 40 | asyncio_pool is 1.53x faster than zoo | 199.72 microseconds |
| bare_thread | 16 | 40 | bare_thread is 1.17x faster than zoo | 199.72 microseconds |
| process_pool | 16 | 40 | process_pool is 1.31x faster than zoo | 199.72 microseconds |
| thread_pool | 16 | 40 | thread_pool is 2.24x faster than zoo | 199.72 microseconds |
| asyncio_pool | 16 | 300 | asyncio_pool is 1.18x faster than zoo | 464.10 microseconds |
| bare_thread | 16 | 300 | bare_thread is 1.11x faster than zoo | 464.10 microseconds |
| process_pool | 16 | 300 | process_pool is 1.67x faster than zoo | 464.10 microseconds |
| thread_pool | 16 | 300 | thread_pool is 1.31x faster than zoo | 464.10 microseconds |
| asyncio_pool | 16 | 2700 | asyncio_pool is 1.04x faster than zoo | 2.869 milliseconds |
| bare_thread | 16 | 2700 | bare_thread is 1.07x faster than zoo | 2.869 milliseconds |
| process_pool | 16 | 2700 | process_pool is 2.03x faster than zoo | 2.869 milliseconds |
| thread_pool | 16 | 2700 | thread_pool is 1.35x faster than zoo | 2.869 milliseconds |
| asyncio_pool | 16 | 10000 | asyncio_pool is 1.02x faster than zoo | 10.104 milliseconds |
| bare_thread | 16 | 10000 | bare_thread is 1.02x faster than zoo | 10.104 milliseconds |
| process_pool | 16 | 10000 | process_pool is 2.15x faster than zoo | 10.104 milliseconds |
| thread_pool | 16 | 10000 | thread_pool is 1.04x faster than zoo | 10.104 milliseconds |
| asyncio_pool | 64 | 40 | asyncio_pool is 1.59x faster than zoo | 195.85 microseconds |
| bare_thread | 64 | 40 | bare_thread is 1.12x faster than zoo | 195.85 microseconds |
| thread_pool | 64 | 40 | thread_pool is 2.16x faster than zoo | 195.85 microseconds |
| asyncio_pool | 64 | 300 | asyncio_pool is 1.22x faster than zoo | 465.14 microseconds |
| bare_thread | 64 | 300 | bare_thread is 1.09x faster than zoo | 465.14 microseconds |
| process_pool | 64 | 300 | process_pool is 1.50x faster than zoo | 465.14 microseconds |
| thread_pool | 64 | 300 | thread_pool is 1.31x faster than zoo | 465.14 microseconds |
| asyncio_pool | 64 | 2700 | asyncio_pool is 1.03x faster than zoo | 2.881 milliseconds |
| bare_thread | 64 | 2700 | bare_thread is 1.04x faster than zoo | 2.881 milliseconds |
| process_pool | 64 | 2700 | process_pool is 2.10x faster than zoo | 2.881 milliseconds |
| thread_pool | 64 | 2700 | thread_pool is 1.04x faster than zoo | 2.881 milliseconds |
| apscheduler | 64 | 10000 | apscheduler is 1.07x faster than zoo | 10.631 milliseconds |
| bare_thread | 64 | 10000 | bare_thread is 1.08x faster than zoo | 10.631 milliseconds |
| process_pool | 64 | 10000 | process_pool is 2.29x faster than zoo | 10.631 milliseconds |

## tiers with no winner

> Tiers where the two sides **cannot be told apart**: the difference is smaller than that tier's band. The band comes from the two sides' own across-round spread (stdev / median of the per-task medians) **within the same run**, so it varies with the sample and the reader can recompute it from the archived per-round samples. It also covers every measured tier and states a reason for each  -  **this section is not a place to hide things**: reporting noise-level differences as '1.00x faster' is.

> criterion: band = the framework's own across-round relative spread plus the baseline's; a single side's across-round relative spread = stdev / median of that unit's per-round per-task medians

band this run: 1.0% to 35.2% (it varies with each tier's spread).

| baseline | concurrency | body tier (microseconds) | ratio | reason |
|---|---|---|---|---|
| asyncio_pool | 1 | 40 | 1.23x | the two sides differ by 22.6%, less than the band of 35.2%  -  no winner |
| bare_thread | 1 | 40 | 0.83x | the two sides differ by 16.8%, less than the band of 19.7%  -  no winner |
| asyncio_pool | 1 | 300 | 0.94x | the two sides differ by 6.4%, less than the band of 8.8%  -  no winner |
| bare_thread | 1 | 300 | 0.92x | the two sides differ by 8.4%, less than the band of 9.5%  -  no winner |
| bare_thread | 4 | 40 | 0.84x | the two sides differ by 15.7%, less than the band of 27.2%  -  no winner |
| process_pool | 4 | 40 | 0.96x | the two sides differ by 4.4%, less than the band of 16.2%  -  no winner |
| asyncio_pool | 4 | 2700 | 0.98x | the two sides differ by 1.8%, less than the band of 3.1%  -  no winner |
| apscheduler | 4 | 10000 | 1.01x | the two sides differ by 0.5%, less than the band of 1.4%  -  no winner |
| apscheduler | 16 | 10000 | 1.01x | the two sides differ by 1.2%, less than the band of 1.7%  -  no winner |
| process_pool | 64 | 40 | 1.02x | the two sides differ by 2.1%, less than the band of 11.5%  -  no winner |
| apscheduler | 64 | 2700 | 1.02x | the two sides differ by 2.0%, less than the band of 3.0%  -  no winner |
| asyncio_pool | 64 | 10000 | 0.98x | the two sides differ by 2.2%, less than the band of 3.8%  -  no winner |
| thread_pool | 64 | 10000 | 0.97x | the two sides differ by 2.8%, less than the band of 3.4%  -  no winner |

## caliber limitations and sources of bias

- **measured layer**: dispatch layer: drives BaseWaiter.execute_service() directly (thread-pool mode) instead of Master's full run loop - the latter's service loop fires once per second, which would lift the latency scale from microseconds to seconds. **One scheduling round per submitted task**, so that whole round's bookkeeping is charged to this single task; Master's usage serves several workers per round, which would spread that cost over tasks
- **scope note** (apscheduler): APScheduler's scheduled-job model, including job registration and trigger-decision cost; the library swallows exceptions raised inside jobs, so this tier re-raises them during drain. **Caliber note**: the library offers no job-completion notification, so the completion signal comes from this tier's Condition, and the cost of one notification per task counts toward its end-to-end
- **scope note** (asyncio_pool): the asyncio event loop dispatches **synchronous** bodies (equivalent to loop.run_in_executor), not native coroutine concurrency; the difference from a thread pool is the event loop's own cost
- **scope note** (bare_thread): one threading.Thread per task, no pooling and no reuse; thread creation cost counts toward end-to-end. Bodies' self-reported timings are collected with a lock-free list.append (atomic under the GIL) - **instrumentation must not charge one tier a cost other tiers do not pay**, or the comparison becomes a comparison of instrumentation
- **scope note** (process_pool): ProcessPoolExecutor: pickling the body and its return value across processes counts toward end-to-end, so this tier is not directly comparable to the multi-threaded tiers
- **scope note** (thread_pool): concurrent.futures.ThreadPoolExecutor, pool size = concurrency; threads are reused, with no task-queue semantics
- **scope note** (zoo): end-to-end stops at the worker completing and being deregistered from the in-flight table; the framework's later result delivery splits by generation: the previous generation's delivery path is broken (dispatch looks reactors up by reactor_name rather than by topic, so every call raises), and the completion signal can only come from the framework's internal callback; the current generation has that fixed, and the completion signal comes from the framework's own result reactor. **Caliber note**: the completion signal comes from this tier's Condition, and the cost of one notification per task counts toward its end-to-end; a bias in that direction makes the framework under test look slower, i.e. it works against it rather than for it
- **groups with no overhead figure**: for the (adapter, concurrency) groups below, concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values and meaningless change factors). This report therefore **gives them no framework overhead and no crossing point**; end-to-end, throughput and the body's self-reported values remain valid and not one field has been dropped from the raw archive. The criterion is the logical core count from the environment self-report, plus one evidential backstop: a computed negative overhead is likewise treated as unreadable  -  a framework can only add time, never remove it
- **native task line: no verdict**: the native task line came out not_applicable (the framework under test has no such capability (symbols not found: adaptive_params, dual_arm_worker, native_params, native_task_worker) - this line does not apply to this version, and that is not a failure) and produced no figures. It is reported here rather than folded into the self-check: one line's failure must not hold back the other versions' data or the main matrix
- **self-check items excluded from the verdict**: for these units the body-tier deviation did not take part in the verdict (concurrency above the lowest, or a tier so short that wall clock is dominated by scheduling granularity). The raw values are still in place, but the wall clock at those two places is not enough to conclude whether the calibration is right  -  **factor this in when reading those tiers' numbers**

> absolute durations: valid only within a single run; do not compare across runs

## bring your own baseline

the baselines of this harness are **open**: implement the four methods of `zoo_bench.adapters.BaseAdapter` (`setup` / `submit` / `drain` / `teardown`), register with `zoo_bench.adapters.registry.register`, or list your module path in `extra_modules` of `matrix.yaml`  -  **no file in this repository needs to change**. Your implementation goes through the same equivalence verification (submit N bodies, assert all ran and each exactly once); failing it means the data is not trusted. **We deliberately write this path into the report**: a report written by the framework's own maintainer, measuring that framework, gives every reason to suspect a slowed baseline or a chosen caliber; letting anyone submit their own baseline is what makes this report challengeable and falsifiable.

## self-check

- overall measurement self-check: passed
- process isolation: {'parent_pid': 29954, 'child_pids': [30410, 30413, 30416, 30419, 30422, 30428, 30434, 30440, 30446, 30464, 30482, 30500, 30521, 30577, 30634, 30692, 30758, 30760, 30762, 30764, 30766, 30771, 30776, 30781, 30786, 30803, 30821, 30838, 30856, 30918, 30977, 31039, 31104, 31138, 31172, 31206, 31240, 31373, 31506, 31639, 31772, 32301, 32830, 33359, 33888, 36001, 38115, 40228, 42343, 42347, 42351, 42355, 42359, 42366, 42373, 42380, 42387, 42406, 42425, 42444, 42463, 42530, 42597, 42664, 42732, 42734, 42736, 42738, 42740, 42745, 42750, 42755, 42760, 42777, 42794, 42811, 42829, 42885, 42941, 43000, 43068, 43070, 43072, 43074, 43076, 43081, 43086, 43091, 43096, 43113, 43130, 43147, 43164, 43205, 43239, 43280], 'all_distinct': True, 'none_equals_parent': True}

## Appendix: all values

> the body carries only pivot tables for readability; this appendix gives **every field as measured before any trimming** per unit (percentiles, spread, measured body durations, ...). **Not one value is dropped**: the details live inside the report body itself, not behind links, folds or attachments  -  'verifiable' that requires a separate lookup is not verifiable

### Appendix: Latency percentiles and jitter

| scheme | concurrency | body tier (microseconds) | median | p95 | p99 | relative spread |
|---|---|---|---|---|---|---|
| apscheduler | 1 | 40 | 281.24 microseconds | 330.91 microseconds | 334.09 microseconds | 0.071 |
| apscheduler | 1 | 300 | 544.52 microseconds | 572.11 microseconds | 578.66 microseconds | 0.035 |
| apscheduler | 1 | 2700 | 2.884 milliseconds | 2.955 milliseconds | 3.076 milliseconds | 0.019 |
| apscheduler | 1 | 10000 | 10.156 milliseconds | 10.294 milliseconds | 10.376 milliseconds | 0.013 |
| apscheduler | 4 | 40 | 255.42 microseconds | 282.19 microseconds | 297.32 microseconds | 0.047 |
| apscheduler | 4 | 300 | 516.26 microseconds | 547.34 microseconds | 563.64 microseconds | 0.029 |
| apscheduler | 4 | 2700 | 2.983 milliseconds | 3.037 milliseconds | 3.077 milliseconds | 0.013 |
| apscheduler | 4 | 10000 | 10.101 milliseconds | 10.215 milliseconds | 10.306 milliseconds | 0.008 |
| apscheduler | 16 | 40 | 246.53 microseconds | 260.75 microseconds | 274.95 microseconds | 0.043 |
| apscheduler | 16 | 300 | 520.81 microseconds | 531.42 microseconds | 533.12 microseconds | 0.021 |
| apscheduler | 16 | 2700 | 2.998 milliseconds | 3.061 milliseconds | 3.068 milliseconds | 0.015 |
| apscheduler | 16 | 10000 | 10.223 milliseconds | 10.325 milliseconds | 10.429 milliseconds | 0.011 |
| apscheduler | 64 | 40 | 245.59 microseconds | 261.27 microseconds | 273.04 microseconds | 0.058 |
| apscheduler | 64 | 300 | 514.29 microseconds | 540.01 microseconds | 549.92 microseconds | 0.031 |
| apscheduler | 64 | 2700 | 2.940 milliseconds | 3.031 milliseconds | 3.147 milliseconds | 0.020 |
| apscheduler | 64 | 10000 | 9.958 milliseconds | 10.331 milliseconds | 10.436 milliseconds | 0.021 |
| asyncio_pool | 1 | 40 | 213.36 microseconds | 294.34 microseconds | 303.06 microseconds | 0.254 |
| asyncio_pool | 1 | 300 | 405.28 microseconds | 432.68 microseconds | 436.27 microseconds | 0.034 |
| asyncio_pool | 1 | 2700 | 2.701 milliseconds | 2.740 milliseconds | 2.742 milliseconds | 0.007 |
| asyncio_pool | 1 | 10000 | 9.700 milliseconds | 9.791 milliseconds | 9.826 milliseconds | 0.007 |
| asyncio_pool | 4 | 40 | 156.05 microseconds | 178.57 microseconds | 186.23 microseconds | 0.081 |
| asyncio_pool | 4 | 300 | 407.59 microseconds | 435.64 microseconds | 443.37 microseconds | 0.030 |
| asyncio_pool | 4 | 2700 | 2.764 milliseconds | 2.847 milliseconds | 2.873 milliseconds | 0.016 |
| asyncio_pool | 4 | 10000 | 9.786 milliseconds | 9.908 milliseconds | 9.942 milliseconds | 0.006 |
| asyncio_pool | 16 | 40 | 130.72 microseconds | 144.29 microseconds | 146.15 microseconds | 0.048 |
| asyncio_pool | 16 | 300 | 392.41 microseconds | 411.25 microseconds | 422.56 microseconds | 0.024 |
| asyncio_pool | 16 | 2700 | 2.756 milliseconds | 2.816 milliseconds | 2.822 milliseconds | 0.012 |
| asyncio_pool | 16 | 10000 | 9.931 milliseconds | 10.004 milliseconds | 10.010 milliseconds | 0.007 |
| asyncio_pool | 64 | 40 | 123.00 microseconds | 126.31 microseconds | 129.42 microseconds | 0.022 |
| asyncio_pool | 64 | 300 | 381.41 microseconds | 391.30 microseconds | 417.56 microseconds | 0.024 |
| asyncio_pool | 64 | 2700 | 2.804 milliseconds | 2.846 milliseconds | 2.857 milliseconds | 0.008 |
| asyncio_pool | 64 | 10000 | 10.397 milliseconds | 10.561 milliseconds | 10.584 milliseconds | 0.018 |
| bare_thread | 1 | 40 | 144.76 microseconds | 175.84 microseconds | 198.18 microseconds | 0.099 |
| bare_thread | 1 | 300 | 396.70 microseconds | 436.06 microseconds | 444.03 microseconds | 0.042 |
| bare_thread | 1 | 2700 | 2.680 milliseconds | 2.726 milliseconds | 2.739 milliseconds | 0.009 |
| bare_thread | 1 | 10000 | 9.548 milliseconds | 9.655 milliseconds | 9.731 milliseconds | 0.007 |
| bare_thread | 4 | 40 | 168.63 microseconds | 252.60 microseconds | 257.71 microseconds | 0.215 |
| bare_thread | 4 | 300 | 412.07 microseconds | 434.78 microseconds | 440.21 microseconds | 0.026 |
| bare_thread | 4 | 2700 | 2.728 milliseconds | 2.756 milliseconds | 2.765 milliseconds | 0.007 |
| bare_thread | 4 | 10000 | 9.829 milliseconds | 9.887 milliseconds | 9.924 milliseconds | 0.004 |
| bare_thread | 16 | 40 | 170.14 microseconds | 176.65 microseconds | 177.51 microseconds | 0.025 |
| bare_thread | 16 | 300 | 419.21 microseconds | 428.16 microseconds | 436.10 microseconds | 0.013 |
| bare_thread | 16 | 2700 | 2.682 milliseconds | 2.699 milliseconds | 2.700 milliseconds | 0.004 |
| bare_thread | 16 | 10000 | 9.895 milliseconds | 9.953 milliseconds | 10.039 milliseconds | 0.004 |
| bare_thread | 64 | 40 | 174.50 microseconds | 206.96 microseconds | 208.44 microseconds | 0.082 |
| bare_thread | 64 | 300 | 426.28 microseconds | 434.04 microseconds | 440.99 microseconds | 0.010 |
| bare_thread | 64 | 2700 | 2.775 milliseconds | 2.925 milliseconds | 2.984 milliseconds | 0.021 |
| bare_thread | 64 | 10000 | 9.824 milliseconds | 9.862 milliseconds | 10.052 milliseconds | 0.006 |
| process_pool | 1 | 40 | 369.75 microseconds | 523.35 microseconds | 702.27 microseconds | 0.221 |
| process_pool | 1 | 300 | 614.59 microseconds | 646.64 microseconds | 657.87 microseconds | 0.031 |
| process_pool | 1 | 2700 | 2.982 milliseconds | 3.078 milliseconds | 3.162 milliseconds | 0.016 |
| process_pool | 1 | 10000 | 10.087 milliseconds | 10.228 milliseconds | 10.606 milliseconds | 0.013 |
| process_pool | 4 | 40 | 191.30 microseconds | 226.31 microseconds | 233.00 microseconds | 0.105 |
| process_pool | 4 | 300 | 341.25 microseconds | 392.67 microseconds | 397.76 microseconds | 0.111 |
| process_pool | 4 | 2700 | 1.359 milliseconds | 1.636 milliseconds | 2.046 milliseconds | 0.120 |
| process_pool | 4 | 10000 | 4.650 milliseconds | 5.143 milliseconds | 5.672 milliseconds | 0.055 |
| process_pool | 16 | 40 | 152.53 microseconds | 195.85 microseconds | 222.74 microseconds | 0.159 |
| process_pool | 16 | 300 | 277.10 microseconds | 325.51 microseconds | 342.76 microseconds | 0.089 |
| process_pool | 16 | 2700 | 1.413 milliseconds | 1.522 milliseconds | 1.531 milliseconds | 0.035 |
| process_pool | 16 | 10000 | 4.698 milliseconds | 4.896 milliseconds | 4.946 milliseconds | 0.019 |
| process_pool | 64 | 40 | 200.04 microseconds | 241.82 microseconds | 253.06 microseconds | 0.095 |
| process_pool | 64 | 300 | 310.51 microseconds | 345.27 microseconds | 355.80 microseconds | 0.059 |
| process_pool | 64 | 2700 | 1.375 milliseconds | 1.448 milliseconds | 1.489 milliseconds | 0.023 |
| process_pool | 64 | 10000 | 4.637 milliseconds | 4.695 milliseconds | 4.742 milliseconds | 0.007 |
| thread_pool | 1 | 40 | 77.28 microseconds | 97.73 microseconds | 131.24 microseconds | 0.178 |
| thread_pool | 1 | 300 | 324.43 microseconds | 343.06 microseconds | 365.55 microseconds | 0.038 |
| thread_pool | 1 | 2700 | 2.658 milliseconds | 2.708 milliseconds | 2.729 milliseconds | 0.009 |
| thread_pool | 1 | 10000 | 9.698 milliseconds | 9.755 milliseconds | 9.835 milliseconds | 0.006 |
| thread_pool | 4 | 40 | 96.24 microseconds | 109.34 microseconds | 114.24 microseconds | 0.064 |
| thread_pool | 4 | 300 | 352.35 microseconds | 369.30 microseconds | 376.29 microseconds | 0.021 |
| thread_pool | 4 | 2700 | 2.711 milliseconds | 2.755 milliseconds | 2.771 milliseconds | 0.011 |
| thread_pool | 4 | 10000 | 9.811 milliseconds | 9.987 milliseconds | 10.010 milliseconds | 0.008 |
| thread_pool | 16 | 40 | 88.97 microseconds | 96.29 microseconds | 96.75 microseconds | 0.062 |
| thread_pool | 16 | 300 | 354.95 microseconds | 365.09 microseconds | 366.11 microseconds | 0.013 |
| thread_pool | 16 | 2700 | 2.133 milliseconds | 2.172 milliseconds | 2.176 milliseconds | 0.012 |
| thread_pool | 16 | 10000 | 9.734 milliseconds | 9.914 milliseconds | 10.042 milliseconds | 0.009 |
| thread_pool | 64 | 40 | 90.65 microseconds | 93.87 microseconds | 102.53 microseconds | 0.035 |
| thread_pool | 64 | 300 | 353.92 microseconds | 414.87 microseconds | 472.73 microseconds | 0.084 |
| thread_pool | 64 | 2700 | 2.774 milliseconds | 2.889 milliseconds | 2.977 milliseconds | 0.019 |
| thread_pool | 64 | 10000 | 10.331 milliseconds | 10.440 milliseconds | 10.460 milliseconds | 0.014 |
| zoo | 1 | 40 | 174.06 microseconds | 211.81 microseconds | 226.31 microseconds | 0.098 |
| zoo | 1 | 300 | 433.22 microseconds | 475.65 microseconds | 481.58 microseconds | 0.054 |
| zoo | 1 | 2700 | 2.800 milliseconds | 2.853 milliseconds | 2.875 milliseconds | 0.009 |
| zoo | 1 | 10000 | 9.885 milliseconds | 9.996 milliseconds | 10.110 milliseconds | 0.007 |
| zoo | 4 | 40 | 200.10 microseconds | 217.10 microseconds | 218.93 microseconds | 0.057 |
| zoo | 4 | 300 | 457.43 microseconds | 477.75 microseconds | 504.06 microseconds | 0.031 |
| zoo | 4 | 2700 | 2.814 milliseconds | 2.893 milliseconds | 2.915 milliseconds | 0.016 |
| zoo | 4 | 10000 | 10.050 milliseconds | 10.125 milliseconds | 10.177 milliseconds | 0.006 |
| zoo | 16 | 40 | 199.72 microseconds | 210.55 microseconds | 214.54 microseconds | 0.026 |
| zoo | 16 | 300 | 464.10 microseconds | 477.83 microseconds | 481.00 microseconds | 0.016 |
| zoo | 16 | 2700 | 2.869 milliseconds | 2.904 milliseconds | 2.919 milliseconds | 0.010 |
| zoo | 16 | 10000 | 10.104 milliseconds | 10.191 milliseconds | 10.304 milliseconds | 0.006 |
| zoo | 64 | 40 | 195.85 microseconds | 203.19 microseconds | 204.54 microseconds | 0.020 |
| zoo | 64 | 300 | 465.14 microseconds | 470.93 microseconds | 474.34 microseconds | 0.009 |
| zoo | 64 | 2700 | 2.881 milliseconds | 2.939 milliseconds | 2.962 milliseconds | 0.010 |
| zoo | 64 | 10000 | 10.631 milliseconds | 10.833 milliseconds | 11.023 milliseconds | 0.020 |

### Appendix: Framework overhead share

| scheme | concurrency | body tier (microseconds) | framework overhead | overhead share | body as measured |
|---|---|---|---|---|---|
| apscheduler | 1 | 40 | 232.64 microseconds | 82.72% | 48.60 microseconds |
| apscheduler | 1 | 300 | 250.38 microseconds | 45.98% | 294.14 microseconds |
| apscheduler | 1 | 2700 | 295.90 microseconds | 10.26% | 2.588 milliseconds |
| apscheduler | 1 | 10000 | 365.37 microseconds | 3.60% | 9.790 milliseconds |
| apscheduler | 4 | 40 | 199.69 microseconds | 78.18% | 55.73 microseconds |
| apscheduler | 4 | 300 | 207.65 microseconds | 40.22% | 308.61 microseconds |
| apscheduler | 4 | 2700 | 291.97 microseconds | 9.79% | 2.691 milliseconds |
| apscheduler | 4 | 10000 | - | - | 14.850 milliseconds |
| apscheduler | 16 | 40 | - | - | 52.28 microseconds |
| apscheduler | 16 | 300 | - | - | 309.42 microseconds |
| apscheduler | 16 | 2700 | - | - | 2.739 milliseconds |
| apscheduler | 16 | 10000 | - | - | 15.698 milliseconds |
| apscheduler | 64 | 40 | - | - | 50.46 microseconds |
| apscheduler | 64 | 300 | - | - | 306.59 microseconds |
| apscheduler | 64 | 2700 | - | - | 2.676 milliseconds |
| apscheduler | 64 | 10000 | - | - | 15.208 milliseconds |
| asyncio_pool | 1 | 40 | 148.75 microseconds | 69.72% | 64.61 microseconds |
| asyncio_pool | 1 | 300 | 107.53 microseconds | 26.53% | 297.76 microseconds |
| asyncio_pool | 1 | 2700 | 123.69 microseconds | 4.58% | 2.578 milliseconds |
| asyncio_pool | 1 | 10000 | 153.03 microseconds | 1.58% | 9.547 milliseconds |
| asyncio_pool | 4 | 40 | 102.43 microseconds | 65.64% | 53.62 microseconds |
| asyncio_pool | 4 | 300 | 104.85 microseconds | 25.72% | 302.74 microseconds |
| asyncio_pool | 4 | 2700 | 111.01 microseconds | 4.02% | 2.652 milliseconds |
| asyncio_pool | 4 | 10000 | - | - | 13.992 milliseconds |
| asyncio_pool | 16 | 40 | - | - | 53.62 microseconds |
| asyncio_pool | 16 | 300 | - | - | 304.98 microseconds |
| asyncio_pool | 16 | 2700 | - | - | 2.632 milliseconds |
| asyncio_pool | 16 | 10000 | - | - | 19.975 milliseconds |
| asyncio_pool | 64 | 40 | - | - | 52.29 microseconds |
| asyncio_pool | 64 | 300 | - | - | 302.33 microseconds |
| asyncio_pool | 64 | 2700 | - | - | 2.675 milliseconds |
| asyncio_pool | 64 | 10000 | - | - | 81.526 milliseconds |
| bare_thread | 1 | 40 | 92.45 microseconds | 63.87% | 52.31 microseconds |
| bare_thread | 1 | 300 | 101.31 microseconds | 25.54% | 295.39 microseconds |
| bare_thread | 1 | 2700 | 114.23 microseconds | 4.26% | 2.566 milliseconds |
| bare_thread | 1 | 10000 | 126.52 microseconds | 1.33% | 9.422 milliseconds |
| bare_thread | 4 | 40 | 116.46 microseconds | 69.06% | 52.17 microseconds |
| bare_thread | 4 | 300 | 118.13 microseconds | 28.67% | 293.94 microseconds |
| bare_thread | 4 | 2700 | 127.88 microseconds | 4.69% | 2.601 milliseconds |
| bare_thread | 4 | 10000 | - | - | 14.814 milliseconds |
| bare_thread | 16 | 40 | - | - | 51.40 microseconds |
| bare_thread | 16 | 300 | - | - | 286.58 microseconds |
| bare_thread | 16 | 2700 | - | - | 2.539 milliseconds |
| bare_thread | 16 | 10000 | - | - | 14.979 milliseconds |
| bare_thread | 64 | 40 | - | - | 51.89 microseconds |
| bare_thread | 64 | 300 | - | - | 296.66 microseconds |
| bare_thread | 64 | 2700 | - | - | 2.640 milliseconds |
| bare_thread | 64 | 10000 | - | - | 15.142 milliseconds |
| process_pool | 1 | 40 | 313.46 microseconds | 84.78% | 56.29 microseconds |
| process_pool | 1 | 300 | 304.04 microseconds | 49.47% | 310.54 microseconds |
| process_pool | 1 | 2700 | 338.24 microseconds | 11.34% | 2.644 milliseconds |
| process_pool | 1 | 10000 | 342.42 microseconds | 3.39% | 9.744 milliseconds |
| process_pool | 4 | 40 | 101.51 microseconds | 53.06% | 89.79 microseconds |
| process_pool | 4 | 300 | - | - | 568.22 microseconds |
| process_pool | 4 | 2700 | - | - | 4.781 milliseconds |
| process_pool | 4 | 10000 | - | - | 17.953 milliseconds |
| process_pool | 16 | 40 | - | - | 91.18 microseconds |
| process_pool | 16 | 300 | - | - | 583.81 microseconds |
| process_pool | 16 | 2700 | - | - | 9.970 milliseconds |
| process_pool | 16 | 10000 | - | - | 52.459 milliseconds |
| process_pool | 64 | 40 | - | - | 91.01 microseconds |
| process_pool | 64 | 300 | - | - | 570.05 microseconds |
| process_pool | 64 | 2700 | - | - | 28.234 milliseconds |
| process_pool | 64 | 10000 | - | - | 159.612 milliseconds |
| thread_pool | 1 | 40 | 30.24 microseconds | 39.13% | 47.04 microseconds |
| thread_pool | 1 | 300 | 36.08 microseconds | 11.12% | 288.35 microseconds |
| thread_pool | 1 | 2700 | 48.20 microseconds | 1.81% | 2.610 milliseconds |
| thread_pool | 1 | 10000 | 70.99 microseconds | 0.73% | 9.627 milliseconds |
| thread_pool | 4 | 40 | 46.57 microseconds | 48.39% | 49.67 microseconds |
| thread_pool | 4 | 300 | 53.93 microseconds | 15.31% | 298.42 microseconds |
| thread_pool | 4 | 2700 | 99.41 microseconds | 3.67% | 2.611 milliseconds |
| thread_pool | 4 | 10000 | - | - | 15.069 milliseconds |
| thread_pool | 16 | 40 | - | - | 48.79 microseconds |
| thread_pool | 16 | 300 | - | - | 305.06 microseconds |
| thread_pool | 16 | 2700 | - | - | 2.035 milliseconds |
| thread_pool | 16 | 10000 | - | - | 19.866 milliseconds |
| thread_pool | 64 | 40 | - | - | 47.76 microseconds |
| thread_pool | 64 | 300 | - | - | 297.37 microseconds |
| thread_pool | 64 | 2700 | - | - | 2.692 milliseconds |
| thread_pool | 64 | 10000 | - | - | 59.527 milliseconds |
| zoo | 1 | 40 | 123.88 microseconds | 71.17% | 50.17 microseconds |
| zoo | 1 | 300 | 131.17 microseconds | 30.28% | 302.05 microseconds |
| zoo | 1 | 2700 | 136.04 microseconds | 4.86% | 2.664 milliseconds |
| zoo | 1 | 10000 | 147.92 microseconds | 1.50% | 9.737 milliseconds |
| zoo | 4 | 40 | 144.96 microseconds | 72.44% | 55.14 microseconds |
| zoo | 4 | 300 | 151.15 microseconds | 33.04% | 306.27 microseconds |
| zoo | 4 | 2700 | 200.65 microseconds | 7.13% | 2.613 milliseconds |
| zoo | 4 | 10000 | - | - | 14.817 milliseconds |
| zoo | 16 | 40 | - | - | 52.89 microseconds |
| zoo | 16 | 300 | - | - | 305.22 microseconds |
| zoo | 16 | 2700 | - | - | 2.658 milliseconds |
| zoo | 16 | 10000 | - | - | 20.089 milliseconds |
| zoo | 64 | 40 | - | - | 52.68 microseconds |
| zoo | 64 | 300 | - | - | 303.54 microseconds |
| zoo | 64 | 2700 | - | - | 2.679 milliseconds |
| zoo | 64 | 10000 | - | - | 51.276 milliseconds |

### Appendix: Throughput and concurrency scaling

| scheme | concurrency | body tier (microseconds) | throughput (tasks/s) |
|---|---|---|---|
| apscheduler | 1 | 40 | 3555.7 |
| apscheduler | 1 | 300 | 1836.5 |
| apscheduler | 1 | 2700 | 346.7 |
| apscheduler | 1 | 10000 | 98.5 |
| apscheduler | 4 | 40 | 15660.3 |
| apscheduler | 4 | 300 | 7748.0 |
| apscheduler | 4 | 2700 | 1340.8 |
| apscheduler | 4 | 10000 | 396.0 |
| apscheduler | 16 | 40 | 64899.7 |
| apscheduler | 16 | 300 | 30721.3 |
| apscheduler | 16 | 2700 | 5337.6 |
| apscheduler | 16 | 10000 | 1565.0 |
| apscheduler | 64 | 40 | 260593.3 |
| apscheduler | 64 | 300 | 124444.1 |
| apscheduler | 64 | 2700 | 21770.5 |
| apscheduler | 64 | 10000 | 6427.2 |
| asyncio_pool | 1 | 40 | 4686.8 |
| asyncio_pool | 1 | 300 | 2467.4 |
| asyncio_pool | 1 | 2700 | 370.2 |
| asyncio_pool | 1 | 10000 | 103.1 |
| asyncio_pool | 4 | 40 | 25632.9 |
| asyncio_pool | 4 | 300 | 9813.8 |
| asyncio_pool | 4 | 2700 | 1447.4 |
| asyncio_pool | 4 | 10000 | 408.8 |
| asyncio_pool | 16 | 40 | 122396.1 |
| asyncio_pool | 16 | 300 | 40773.2 |
| asyncio_pool | 16 | 2700 | 5805.4 |
| asyncio_pool | 16 | 10000 | 1611.1 |
| asyncio_pool | 64 | 40 | 520342.3 |
| asyncio_pool | 64 | 300 | 167797.3 |
| asyncio_pool | 64 | 2700 | 22823.4 |
| asyncio_pool | 64 | 10000 | 6155.5 |
| bare_thread | 1 | 40 | 6907.9 |
| bare_thread | 1 | 300 | 2520.8 |
| bare_thread | 1 | 2700 | 373.2 |
| bare_thread | 1 | 10000 | 104.7 |
| bare_thread | 4 | 40 | 23720.3 |
| bare_thread | 4 | 300 | 9707.1 |
| bare_thread | 4 | 2700 | 1466.1 |
| bare_thread | 4 | 10000 | 407.0 |
| bare_thread | 16 | 40 | 94038.7 |
| bare_thread | 16 | 300 | 38167.2 |
| bare_thread | 16 | 2700 | 5965.9 |
| bare_thread | 16 | 10000 | 1617.0 |
| bare_thread | 64 | 40 | 366766.2 |
| bare_thread | 64 | 300 | 150135.6 |
| bare_thread | 64 | 2700 | 23067.0 |
| bare_thread | 64 | 10000 | 6514.5 |
| process_pool | 1 | 40 | 2704.5 |
| process_pool | 1 | 300 | 1627.1 |
| process_pool | 1 | 2700 | 335.3 |
| process_pool | 1 | 10000 | 99.1 |
| process_pool | 4 | 40 | 20909.5 |
| process_pool | 4 | 300 | 11721.5 |
| process_pool | 4 | 2700 | 2944.3 |
| process_pool | 4 | 10000 | 860.2 |
| process_pool | 16 | 40 | 104898.1 |
| process_pool | 16 | 300 | 57740.1 |
| process_pool | 16 | 2700 | 11327.3 |
| process_pool | 16 | 10000 | 3406.0 |
| process_pool | 64 | 40 | 319932.6 |
| process_pool | 64 | 300 | 206112.0 |
| process_pool | 64 | 2700 | 46544.6 |
| process_pool | 64 | 10000 | 13803.0 |
| thread_pool | 1 | 40 | 12939.1 |
| thread_pool | 1 | 300 | 3082.3 |
| thread_pool | 1 | 2700 | 376.2 |
| thread_pool | 1 | 10000 | 103.1 |
| thread_pool | 4 | 40 | 41562.7 |
| thread_pool | 4 | 300 | 11352.2 |
| thread_pool | 4 | 2700 | 1475.5 |
| thread_pool | 4 | 10000 | 407.7 |
| thread_pool | 16 | 40 | 179841.5 |
| thread_pool | 16 | 300 | 45076.5 |
| thread_pool | 16 | 2700 | 7501.0 |
| thread_pool | 16 | 10000 | 1643.7 |
| thread_pool | 64 | 40 | 705985.9 |
| thread_pool | 64 | 300 | 180832.2 |
| thread_pool | 64 | 2700 | 23074.4 |
| thread_pool | 64 | 10000 | 6194.7 |
| zoo | 1 | 40 | 5745.3 |
| zoo | 1 | 300 | 2308.3 |
| zoo | 1 | 2700 | 357.2 |
| zoo | 1 | 10000 | 101.2 |
| zoo | 4 | 40 | 19989.9 |
| zoo | 4 | 300 | 8744.6 |
| zoo | 4 | 2700 | 1421.5 |
| zoo | 4 | 10000 | 398.0 |
| zoo | 16 | 40 | 80111.5 |
| zoo | 16 | 300 | 34475.3 |
| zoo | 16 | 2700 | 5576.0 |
| zoo | 16 | 10000 | 1583.6 |
| zoo | 64 | 40 | 326787.3 |
| zoo | 64 | 300 | 137591.9 |
| zoo | 64 | 2700 | 22213.7 |
| zoo | 64 | 10000 | 6020.1 |