As Mike mentioned, the design doc <https://docs.google.com/document/d/1xaoF9iSOojrlPrHZaKIJMK4iRZKA3AD6pQvbSy4ueUQ/edit?usp=sharing> links to the spreadsheet <https://docs.google.com/spreadsheets/d/1hW-xXj9zEdkTJMVE4b4qcflB00WxsOG72Y0LVtSjcWI/edit?usp=sharing> with the full list of candidate URLs that should be considered IF this ever gets as far as shipping (as of February) as well as the query to generate the candidate list from the public HTTP Archive dataset for any month after January 2025 (when the response body hashes were added to the HTTP Archive dataset). There are scripts from a lot of different companies (Facebook, Microsoft, Wix, Wordpress, Yandex, Amazon, etc). The ones called out in the design doc are specific for running an experiment to gather data and see if there is any measurable benefit for even the most-pervasive resources from a partner that is well positioned to be able to measure the impact of something like a 1% Chrome experiment.
>From the browser-side we're only likely to see small movement in some low-level metrics (like cache hit rate and bandwidth) but it's not likely to move LCP or other user-facing metrics that we observe but that doesn't mean there's no impact to the web (users, publishers and service providers) which is why we need a partner that can help fill in those details. The partitioned cache was introduced in 2020 <https://developer.chrome.com/blog/http-cache-partitioning> with the privacy benefits outweighing any potential performance impacts. This experiment hopes to quantify those impacts and see if there is a path to restoring the benefits while maintaining the privacy protections that came with the split cache. If there is no impact (or it isn't enough to justify the complexity) then that's pretty much the end of the story for most of the efforts around shared caching (it keeps popping up in different contexts). It's not guaranteed that there will be significant benefits - it's not unusual for the disk cache to be slower than the network, even today. If there IS an impact, then we can use the data from the result to weigh the benefits to the platform against the added complexity and determine what a path forward should look like (if there is one). It's probably important to note that this is specifically only the HTTP disk cache for the request responses. The v8 code cache (where there are substantial benefits) continues to remain partitioned for security reasons (and was partitioned long before the HTTP cache). On Wed, Apr 30, 2025 at 5:07 AM Jxck <j...@chromium.org> wrote: > Experiment candidates in Design docs only includes > > - google-analytics.com > - googlesyndication.com > - doubleclick.net > - googletagmanager.com > - googleapis.com > - youtube.com > - etc > > Even if this list were based on a neutral data analysis, it can't help > thinking that Chrome is intended to ship changes that are only benefits for > Google's business. > On Wednesday, April 30, 2025 at 6:02:20 AM UTC+9 danielher...@gmail.com > wrote: > >> Personally I'm not a fan of this idea of giving special performance >> boosts to big properties like YouTube or Google Ads. That's creating an >> unfair advantage over smaller players in the market. And I have to point >> out that Google is currently under antitrust scrutiny for allegedly doing >> such. >> >> On Tue, Apr 29, 2025 at 2:20 PM Patrick Meenan <pmee...@chromium.org> >> wrote: >> >>> The previous experiment required an exact match of a full URL and >>> payload hash (on a path to automate allow-listing responses). >>> Unfortunately, most of the interesting third-party resources are updated >>> quite regularly (usually at least daily if not every few hours) so the URL >>> list and payload hashes were both very stale by the time the experiment was >>> run. >>> >>> This experiment learned from those results and is targeting a MUCH >>> smaller list of manually curated resources that are much more popular and >>> uses a URLPattern to allow for versioned variants of the URLs (for things >>> like the YT player JS or pubads_impl which include versions in the path) >>> and removes the hash-matching requirement for the payload (for things like >>> analytics.js which update in-place). >>> >>> It won't impact as many third-party resources, by design, but the >>> resources that it does impact are going to be MUCH more likely to be seen >>> by most users multiple times within a browsing session (and within the >>> update window for any of them). >>> >>> On Tue, Apr 29, 2025 at 10:52 AM Mike Taylor <miketa...@chromium.org> >>> wrote: >>> >>>> Hey Patrick, >>>> >>>> This sounds related to a previous experiment ( >>>> https://groups.google.com/a/chromium.org/g/blink-dev/c/9xWJK3IgJb4/m/SYLfUccOBgAJ) >>>> - could you describe how this experiment is similar or different (or >>>> perhaps builds upon the results from the previous experiment)? >>>> >>>> thanks, >>>> Mike >>>> On 4/23/25 10:46 AM, Patrick Meenan wrote: >>>> >>>> Contact emails pmee...@chromium.org >>>> >>>> Specification None >>>> >>>> Design docs >>>> >>>> https://docs.google.com/document/d/1xaoF9iSOojrlPrHZaKIJMK4iRZKA3AD6pQvbSy4ueUQ/edit?usp=sharing >>>> >>>> Summary >>>> >>>> For a small number (hundreds) of hand-curated static third-party >>>> scripts that are used on a large portion of the web, Chrome will use a >>>> single-keyed HTTP cache to store those resources. This helps users and site >>>> owners with faster performance for those scripts that are very widely used >>>> while maintaining the privacy protections of the partitioned disk cache. >>>> This feature targets the scripts that most users are likely to see multiple >>>> times across multiple sites in any given browsing session. They are usually >>>> not in the critical path of the page loading and may not impact the common >>>> performance metrics but they are still important for the healthy operation >>>> of the web. The list of candidate scripts is curated from the HTTP Archive >>>> dataset. This includes site-independent things like common analytics >>>> scripts, social media embeds, video player embeds, captcha providers and >>>> ads libraries. It allows for code that uses versioned URLs as long as the >>>> versioning is not a manual process by embedders and that the same version >>>> is sent to everybody at a given point in time with the same contents. This >>>> does not include things like common Javascript libraries where they are >>>> commonly self-hosted or where the URL references a specific version of the >>>> library and it is up to site owners to manually select a version. >>>> >>>> Blink component Blink>Network >>>> <https://issues.chromium.org/issues?q=customfield1222907:%22Blink%3ENetwork%22> >>>> >>>> TAG review None >>>> >>>> TAG review status Not applicable >>>> >>>> Risks >>>> >>>> >>>> Interoperability and Compatibility >>>> >>>> This change is internal to Chrome and should be completely transparent >>>> to the web platform with no interoperability risks. >>>> >>>> >>>> *Gecko*: N/A >>>> >>>> *WebKit*: N/A >>>> >>>> *Web developers*: No signals >>>> >>>> *Other signals*: >>>> >>>> Ergonomics >>>> >>>> N/A >>>> >>>> >>>> Activation >>>> >>>> N/A >>>> >>>> >>>> Security >>>> >>>> Cache partitioning added a level of privacy protection that is being >>>> disabled for a small number of resources where it is deemed safe to do so. >>>> The linked document and issue provide the details on the protections that >>>> are in place to minimize the privacy exposure. >>>> >>>> >>>> WebView application risks >>>> >>>> Does this intent deprecate or change behavior of existing APIs, such >>>> that it has potentially high risk for Android WebView-based applications? >>>> >>>> None >>>> >>>> >>>> Goals for experimentation For the purposes of the experiment, we will >>>> target around a dozen of the most-impactful scripts (mostly ads and >>>> analytics) and work with the script owners to measure the resulting impact >>>> (on performance, reliability and business metrics). >>>> >>>> If these scripts show minimal impact from using a shared cache then >>>> there is no value in the added complexity for the feature and it can >>>> be documented and closed. >>>> >>>> If there is significant impact then it is likely worth pursuing further >>>> (and will also be documented for other similar efforts). >>>> >>>> Ongoing technical constraints >>>> >>>> None >>>> >>>> >>>> Debuggability >>>> >>>> The cache key used for any given resource is exposed in the netlog >>>> <https://www.chromium.org/for-testers/providing-network-details/> >>>> which can be used to verify if a given resource used the shared cache or >>>> not. >>>> >>>> >>>> Will this feature be supported on all six Blink platforms (Windows, >>>> Mac, Linux, ChromeOS, Android, and Android WebView)? Yes >>>> >>>> Is this feature fully tested by web-platform-tests >>>> <https://chromium.googlesource.com/chromium/src/+/main/docs/testing/web_platform_tests.md> >>>> ? No >>>> >>>> Flag name on about://flags None >>>> >>>> Finch feature name CacheSharingForPervasiveScripts >>>> >>>> Non-finch justification None >>>> >>>> Requires code in //chrome? False >>>> >>>> Tracking bug https://issues.chromium.org/u/1/issues/404196743 >>>> >>>> Measurement The success of this feature will be measured directly with >>>> the owners of a small number of targeted scripts with a web-exposed >>>> experiment. >>>> >>>> Estimated milestones >>>> >>>> No milestones specified >>>> >>>> >>>> Link to entry on the Chrome Platform Status >>>> https://chromestatus.com/feature/5202380930678784 >>>> >>>> This intent message was generated by Chrome Platform Status >>>> <https://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 visit >>>> https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAPq58w50QC8Vj9qUT_eyAdz2nFx2jEW0033KFeDeM6v%3D4J-mKw%40mail.gmail.com >>>> <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAPq58w50QC8Vj9qUT_eyAdz2nFx2jEW0033KFeDeM6v%3D4J-mKw%40mail.gmail.com?utm_medium=email&utm_source=footer> >>>> . >>>> >>>> -- >>> 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 visit >>> https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAPq58w5V%3DByzejfvmvPtWy_xMdQ39SGS-vBwAOq3HVtA2CCEng%40mail.gmail.com >>> <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAPq58w5V%3DByzejfvmvPtWy_xMdQ39SGS-vBwAOq3HVtA2CCEng%40mail.gmail.com?utm_medium=email&utm_source=footer> >>> . >>> >> -- 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 visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAPq58w4gw8pVXTdhRvtUuoX8Q3WJ5kSxm_6J1St2G78HQs%2BDBQ%40mail.gmail.com.
