[Webkit-unassigned] [Bug 147824] NetworkProcess: DNS prefetch happens in the Web Process

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=147824

--- Comment #16 from Carlos Garcia Campos <cgarcia at igalia.com> ---
(In reply to comment #15)
> Comment on attachment 262440 [details]
> Updated patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=262440&action=review
> 
> > Source/WebKit2/NetworkProcess/NetworkProcess.h:197
> > +    HashSet<String> m_dnsPrefetchedHosts;
> 
> This HashSet grows indefinitely, and NetworkProcess can exist for weeks or
> even for months.

I thought about it, my initial idea was to cache only in the web process, since the real performance problem is likely when the same web process requests the same hosts a lot of times, rather than different web processes requesting the same host. So we could simplify it by only caching in the web process and assuming that we might end up requesting the same host by two different web processes.

> > Source/WebKit2/WebProcess/WebProcess.h:368
> > +    HashSet<String> m_dnsPrefetchedHosts;
> 
> Is it OK to keep hosts here for a long time? I think that this may entirely
> disable the optimization of prefetching link hosts on hover - the hosts were
> already prefetched when loading, but the OS resolver may have well
> invalidated the cache already.

Good point. Maybe we should only ensure we don't ask the same hosts in a short period of time, instead. For example when hovering, we start a lot of requests for the same link when the mouse is moving.

> DNS prefetching is a really weird thing. We don't have any way to measure
> its effectiveness, which is not how we usually go about performance
> optimizations.

Yeah, indeed.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.webkit.org/pipermail/webkit-unassigned/attachments/20151006/9c9f5aa7/attachment.html>