Open Bug 580117 Opened 14 years ago Updated 2 years ago

Prefetch DNS for links that the mouse gets close to

Categories

(Core :: DOM: Core & HTML, defect, P5)

x86_64
Linux
defect

Tracking

()

People

(Reporter: jduell.mcbugs, Unassigned)

References

Details

In this video

   http://google-chrome-browser.com/why-google-chrome-fast-spotlight-dns-pre-resolution-0

The chromium guys explain that they prefetch DNS for any link that the mouse comes close to.  They say it takes people an average of 200ms to actually get the mouse right over a link they want to click, and in the meantime DNS can be resolving.

Note that we (and they) already prefetch DNS for all links on a page as we parse the HTML.  So the DNS ought to be resolved anyway.  But they claim that all the prefetching going on (multiple tabs, etc.?) can wind up dropping prefetched entries out of the OS's DNS cache, so they do the mouse-hover prefetches, too.

Not likely to be a huge win, but could be worth doing if it's not too much trouble (it should just be a matter of using nsHTMLDNSPrefetch).
it might also be interesting if configuring this was separable from whole-page prefetch.. that would actually cut down on overall prefetch traffic but maybe work as well on the stuff you are interested in.
There's an obvious privacy issue here -- Alice probably doesn't expect that the browser will leak information about which links she moves her mouse close to.

I'm not sure if the privacy situation is improved or not if we also have whole-page prefetching.  Without whole-page prefetching, it's more obvious when a DNS call is caused by mouse movements.  OTOH with whole-page prefetching, we leak more about the contents of the page you're at, which I guess is only a problem if the page is encrypted.
I for one hover over links to see where they point.  I don't always click them, and would be surprised to learn about any predictive fetching at all based on me pointing.

The delta in privacy concern is less since we do whole-page prefetching, but it's still a delta and potentially interesting.

Do we have a grasp on what change in behavior this actually has in practice?  For example, how frequently would the link-pointing cause duplicate DNS resolutions?  On what % of link clicks is the DNS record pushed out due to cache replacement? Is there a way to order or prioritize lookups as to avoid filling the cache with hostnames for links unlikely to be clicked?
https://bugzilla.mozilla.org/show_bug.cgi?id=1472046

Move all DOM bugs that haven’t been updated in more than 3 years and has no one currently assigned to P5.

If you have questions, please contact :mdaly.
Priority: -- → P5
Component: DOM → DOM: Core & HTML
Severity: normal → S3
You need to log in before you can comment on or make changes to this bug.