Thanks! In that case, LGTM to use origin trials for this otherwise web-invisible experiment. :)
-mike On Tue 5. Oct 2021 at 09:51 Andreas Haas <ah...@google.com> wrote: > Hi Mike, > > We want to use the Origin Trial to get some early information from > partners, but then plan to ship with a finch trial and roll it out step by > step while observing the impact on webpages we are not directly aware of. > > Cheers, > Andreas > > On Mon, Oct 4, 2021 at 6:58 PM Mike West <mk...@chromium.org> wrote: > >> If this doesn’t have web-visible behavioral impact, I don’t think you >> generally need API owner approval. In this case, I think it’s reasonable to >> ask for such approval, as you want to use Origin Trials to more clearly >> involve partners in the experimentation process, do their own A/B testing, >> etc. >> >> I think a marginally-longer-than-usual experiment lifetime is fine here, >> since there’s zero burn-in risk (again, because there’s no behavioral >> change). >> >> I am, though, a little concerned about the validity of any data you >> obtain via this self-selected mechanism. Will you be running a more-typical >> percentage trial at the same time to evaluate the impact of the change more >> broadly? >> >> -mike >> >> On Mon 4. Oct 2021 at 18:08 Andreas Haas <ah...@google.com> wrote: >> >>> >>> >>> On Mon, Oct 4, 2021 at 5:39 PM Mike West <mk...@chromium.org> wrote: >>> >>>> On Mon, Oct 4, 2021 at 10:04 AM 'Andreas Haas' via blink-dev < >>>> blink-dev@chromium.org> wrote: >>>> >>>>> Contact emailsah...@chromium.org >>>>> >>>>> ExplainerAt the moment, V8 compiles WebAssembly modules by first >>>>> compiling all functions with the baseline compiler Liftoff, and then >>>>> immediately compiling all functions with the optimizing compiler TurboFan, >>>>> see https://v8.dev/docs/wasm-compilation-pipeline. On some websites >>>>> we see that TurboFan compilation can block JavaScript workers from >>>>> execution. With WebAssembly dynamic tiering we want to reduce the CPU time >>>>> of TurboFan compilation by only optimizing functions which were already >>>>> executed a few times. >>>>> >>>>> SpecificationThis feature is just a new implementation of an already >>>>> implemented specification. >>>>> >>>>> Summary >>>>> >>>>> With WebAssembly Dynamic Tiering, an heuristic decides which functions >>>>> of a WebAssembly module get optimized, and when the optimization is >>>>> triggered. This is an improvement to the existing eager optimization >>>>> approach, where all functions get optimized immediately after baseline >>>>> compilation is finished. WebAssembly Dynamic Tiering reduces the resource >>>>> consumption of the optimizing compiler, and prevents the compiler from >>>>> competing with the web application for resources. >>>>> >>>>> >>>>> Blink componentBlink>JavaScript>WebAssembly >>>>> <https://bugs.chromium.org/p/chromium/issues/list?q=component:Blink%3EJavaScript%3EWebAssembly> >>>>> >>>>> TAG reviewNot applicable >>>>> >>>>> TAG review statusNot applicable >>>>> >>>>> Risks >>>>> WebAssembly dynamic tiering may lead to unexpected performance >>>>> regressions. >>>>> >>>>> Interoperability and Compatibility >>>>> >>>>> >>>>> >>>>> Gecko: No signal >>>>> >>>>> WebKit: No signal >>>>> >>>>> Web developers: No signals >>>>> >>>> >>>> My understanding is that this doesn't create any web-facing change in >>>> behavior, but only a potential change in performance? >>>> >>>> In that case, then signals and TAG review and etc. all seem quite >>>> reasonable to skip. :) >>>> >>>> >>>>> >>>>> >>>>> Goals for experimentation >>>>> >>>>> The goal of the experiment is to allow important partners to >>>>> experiment with the performance impact of WebAssembly dynamic tiering. >>>>> This >>>>> feature may change the startup behavior of WebAssembly code significantly, >>>>> which is why we would like to experiment in the guarded environment of an >>>>> origin trial first. >>>>> >>>>> Reason this experiment is being extended >>>>> >>>>> >>>>> >>>>> Ongoing technical constraints >>>>> >>>>> >>>>> >>>>> Debuggability >>>>> >>>>> Debugging behavior does not change >>>>> >>>>> Will this feature be supported on all six Blink platforms (Windows, >>>>> Mac, Linux, Chrome OS, Android, and Android WebView)?Yes >>>>> >>>>> Is this feature fully tested by web-platform-tests >>>>> <https://chromium.googlesource.com/chromium/src/+/master/docs/testing/web_platform_tests.md> >>>>> ?Yes >>>>> >>>>> Flag nameWebAssemblyDynamicTiering >>>>> >>>>> Requires code in //chrome?False >>>>> >>>>> Estimated milestones >>>>> >>>>> No milestones specified >>>>> >>>> >>>> Can you specify milestones? :) How many releases do you need in order >>>> to evaluate the impact? >>>> >>>> We would like to do the experiment from M96 until M102. We would like >>> to improve the heuristics for when to optimize over time, which requires >>> the longer time span. >>> >>> >>>> -mike >>>> >>>> >>>>> Link to entry on the Chrome Platform Status >>>>> https://chromestatus.com/feature/5685307493056512 >>>>> >>>>> This intent message was generated by Chrome Platform Status >>>>> <https://www.chromestatus.com/>. >>>>> >>>>> -- >>>>> You received this message because you are subscribed to the Google >>>>> Groups "blink-dev" group. >>>>> To unsubscribe from this group and stop receiving emails from it, send >>>>> an email to blink-dev+unsubscr...@chromium.org. >>>>> To view this discussion on the web visit >>>>> https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAELSTveTudJkMbuBMyZ%2BZTv334audVik78gEJTzmjym4X6wJTg%40mail.gmail.com >>>>> <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAELSTveTudJkMbuBMyZ%2BZTv334audVik78gEJTzmjym4X6wJTg%40mail.gmail.com?utm_medium=email&utm_source=footer> >>>>> . >>>>> >>>> -- >> -mike >> > -- -mike -- You received this message because you are subscribed to the Google Groups "blink-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to blink-dev+unsubscr...@chromium.org. To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAKXHy%3DfTE2205jtX9W-NYrsE4RUtJ8NcDYaUTAxobgtxaP2j2g%40mail.gmail.com.
