AUS-1012: Publish the benchmarks from the cpu-performance job, regenerated from data

.github/workflows/build_pull_request.yml

@@ -46,7 +46,7 @@ jobs:
 
   # The measurement: ProcessAsync must scale from 1 to 4 workers. A process-wide serialization point holds the ratio
   # near 1.3 on any core count. Diagnostic on the shared runner (its core count and isolation are not promised, so
-  # this job is not required and continues on error); the release gate is `--scale 3 16 4.0` on the reference machine.
+  # this job is not required and continues on error); the release gate is `--scale 3 32 4.0` in the cpu-performance publish job.
   # The per-serializer diagnostics after the gate (one run per cell, never part of the exit code) go to their own summary section.
   scaling:
     runs-on: ubuntu-latest

AustinHarris.JsonRpc.AspNetCore/README.md

@@ -161,9 +161,9 @@ invoked.
   Replies already finished are flushed before the connection waits on a slow method. When the connection closes,
   the handler waits for the running method to finish and discards its response.
 - **Cost:** every document then goes through `ProcessAsync`. With methods that complete inline, the TCP row measured
-  15.0 M to 15.4 M against 14.3 M to 16.5 M for the synchronous mode on the same day, inside its spread. A method that
+  17 M to 17.1 M against 18.5 M for the synchronous mode in the same benchmark job, about 8 % lower by the medians. A method that
   suspends pays for its own async state, the library's completion state and a continuation per request; the main README's
-  Async table measured 559 B per request for the yielding None row in process at one worker, including the service's own
+  Async table measured 548 B per request for the yielding None row in process at one worker, including the service's own
   allocations. The main README's Kestrel table has both rows, measured with `TestServer_Console --kestrel 3 async`.
 
 A method receives the token by declaring a `[JsonRpcCancellation] CancellationToken` parameter; see

CHANGELOG.md

@@ -50,6 +50,7 @@ behaviour: a breaking change to either means a new major version.
 - Notifications never get a wire response, whatever their outcome.
 - Dates and non-finite numbers are written the same way by every serializer (fraction only when non-zero, `Z`/offset/nothing by `Kind`; `NaN` and the infinities as quoted strings).
 - Clarified that the WebAssembly sample is not a Native AOT or full-trimming compatibility guarantee.
+- The published benchmarks are measured on a Hugging Face Jobs `cpu-performance` host (AMD EPYC 7R13, 32 cores) and regenerated from the job's data (`benchmarks/hf/`, `ingest.py`, `render.py`); the desktop figures are gone from the README except the browser table.
 
 ### Deprecated
 

Json-Rpc/README.md

@@ -135,22 +135,26 @@ The synchronous entry points do not await those methods.
 
 ## Performance
 
-These results compare 1.2.3 and 2.0 with the same five requests on the same machine in one session.
-The 2.0 runs used the built-in serializer on an AMD Ryzen 7 7800X3D with .NET 10, Release and Server GC.
+These results compare 1.2.3 and 2.0 with the same five requests on the same machine in one benchmark job.
+The 2.0 runs used the built-in serializer on a Hugging Face Jobs `cpu-performance` host (AMD EPYC 7R13, 32 cores) with .NET 10, Release and Server GC.
+Each range is the low and high over the job's runs; the ratios are of the medians.
 
 ![JSON-RPC.Net 1.2.3 and 2.0 throughput through the string and byte entry points](https://raw.githubusercontent.com/Astn/JSON-RPC.NET/master/benchmarks/charts/headline-1x-vs-2.svg)
 
+<!-- benchmarks:headline -->
 | Path | RPC/s | Against 1.2.3 |
 | --- | ---: | ---: |
-| 1.2.3, `Task<string> Process(string)`, thread pool, best batch size | 3.08 M | |
-| 2.0, the same string API and the same loop | 13.3 M | 4.3× |
-| 2.0, `Process(bytes)`, 16 dedicated threads | 31.7 M | 10.3× |
-| 2.0, `ProcessAsync(bytes)`, 16 awaited workers | 32.1 M | 10.4× |
+| 1.2.3, `Task<string> Process(string)`, thread pool, best batch size | 1.29 M to 2.41 M | |
+| 2.0, the same string API and the same loop | 9.83 M to 11.1 M | 6.0× |
+| 2.0, `Process(bytes)`, 32 dedicated threads | 44.2 M to 45.7 M | 24.3× |
+| 2.0, `ProcessAsync(bytes)`, 32 awaited workers | 36.3 M to 40.5 M | 20.7× |
+
+<!-- /benchmarks:headline -->
 
 These measurements cover the library without a transport.
 The asynchronous byte row uses methods that complete inline.
 For Kestrel TCP with 256 requests in flight per connection and `EnableAsyncMethods = false`,
-the measured range is 14.3 M to 16.5 M RPC/s on loopback, with clients and server on the same machine.
+the measured figure is 18.5 M RPC/s on loopback, with clients and server on the same machine.
 
 See the [benchmark tables](https://astn.github.io/JSON-RPC.NET/#benchmarks) for conditions
 and the [benchmark explorer](https://astn.github.io/JSON-RPC.NET/benchmarks/charts/explorer.html) for the data.