Re: [w3ctag/design-reviews] Updated review of WebNN API (Issue #933)

@torgo, a challenge here is that WebNN supports more than just GPU compute. What @RafaelCintron mentioned makes sense as a concrete mitigation when the `MLContext` is configured to prefer GPU compute and we need to coordinate with the browser's WebGPU engine anyways for interop purposes (e.g. passing buffers between WebNN graphs and WebGPU shaders). We have the most implementation experience with that scenario when using DirectML on Windows but are actively prototyping with other frameworks such as Core ML on macOS.

When using the CPU or a dedicated ML accelerator the types of potential resource contention and their mitigations are different. I think a general statement similar to WebGPU's reference to [denial of service attacks](https://www.w3.org/TR/webgpu/#security-dos) makes sense to add to WebNN as well, with the understanding that exactly how the mitigations work will be implementation- and configuration-dependent. Implementations should use whatever mechanisms are available from the platform (such as the watchdogs mentioned by WebGPU) to prevent sites from using an unfair amount of system resources but in the end these are shared resources and the use of any compute API will affect overall performance on a fully-loaded system.

-- 
Reply to this email directly or view it on GitHub:
https://github.com/w3ctag/design-reviews/issues/933#issuecomment-2159340815
You are receiving this because you are subscribed to this thread.

Message ID: <w3ctag/design-reviews/issues/933/2159340815@github.com>