This repo tests the message passing speed between the main thread and a worker thread for each of the following situations as available:
-
strings:
Worker.postMessage(<String>) -
json transfer/structured cloning:
Worker.postMessage(<Object>) -
transferable objects:
Worker.postMessage(<Object>, [transferables]) -
channel strings:
channel.postMessage(<String>) -
channel json transfer/structured cloning:
channel.postMessage(<Object>) -
channel transferable objects:
channel.postMessage(<Object>, [transferables])
Each test is run with 2 data setups.
Also note, many data formats can be used only when not using JSON.stringify unless
a bespoke implementation of stringify is created. For instance, stringify misses out
in these areas:
- duplicating/transferring RegExp objects.
- duplicating/transferring Blob, File, and FileList objects.
- duplicating/transferring ImageData objects. The dimensions of the clone's CanvasPixelArray will match the original and have a duplicate of the same pixel data.
- duplicating/transferring objects containing cyclic graphs of references.
This means that even in cases where using JSON.stringify is marginally faster, once these
cases are accounted for it's likely to be slower.
Each test measures round trips: e.g. the time it takes to serialize, send, receive a response, and parse the response. This means this test is biased towards operations that expect a response, while many good setups communicate only worker to main thread. A fair test would test the main thread overhead of processing a response.