On 1/21/2015 1:10 PM, Nicholas Hurley wrote:
So for a quick answer to your question, yes, there is a frecency-like
algorithm for this. It's even tuneable (to an extent) via prefs! (I'm
working on a blog post about this feature that will go into more detail
on this, keep an eye on Planet Mozilla, hopefully I'll find the cycles
to finish it within the next week or so.)
Cool! I'll certainly take a look once it's up!
My concern is that by going to a site (yahoo.com, in your example),
I'm not saying that I currently "trust" any of the other sites linked
off there, even if I've been there before. If I trust them I'll go to
them.
So perhaps we're talking past each other (always a possibility!), but my
understanding of this sentence indicates to me that I've not done a good
job of explaining to you what exactly it is this feature does. My
understanding is that you think we're opening connections to all the
links on a page, or perhaps going even further and opening connections
to hosts used by the pages linked to. We aren't doing that at all. (Note
that we *do* make DNS requests for every link on a page, but that is
independent of this feature, and has existed for much longer than this
feature.) What we *are* doing is opening connections to hosts that serve
content that is required to fully load a page you're visiting. For
example, lots of sites host the HTML on their own domain, but host
things like js, css, and images on a CDN somewhere (with a different
domain name, or a set of sharded domains names). What we're doing is
(effectively) keeping track of what CDNs we loaded stuff from when you
visited yahoo.com (to keep the example), and then the next time you go
to yahoo.com, we'll open connections to those CDNs before we have the
full HTML parsed (or do DNS lookups for those CDNs, or do nothing,
depending on what the frecency algorithm decides is the Right Thing).
Does that explain it better? (Or, alternately, did you already
understand that, and I totally misread what you were saying, and I just
wasted a couple minutes and a few hundred bytes typing stuff you already
understood? :-) )
Thanks for explaining in more detail. This is not what I thought you were talking about *at all*. I was under the
impression that some sort of prefetching was being done on *links* on the page (where "prefetching" is limited
to opening SSL connections/doing DNS requests). Sorry for the run around then, perhaps I mistook "links"
somewhere to mean <a href="">...</a> where it meant something else. This sounds like a much more
reasonable feature to me!
--Patrick
_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform
