Le 07/08/2013 00:34, Jack Moffitt a écrit :
Love it! This is a game changer for web development. I mean it.
Does this mean you are no longer worried about that suggested benchmark? :)
:-p
I didn't answer, but Patrick Walton made a good point on the fact that
how people use the technology depends on its performance.
But I still believe it's too early to optimize for 100 iframes. (note
that it's just my opinion).
We're trying to find some early benchmarks, so this kind of feedback
is valuable. I figured I'd try and close the loop here to see if this
addresses your concern or if perhaps with the extra explanation you
have an even better idea for us.
If that was all just up to me (it isn't) and assuming resources aren't a
constraint, I would focus, both for compat and performance on top
websites (maybe top Alexa 10-15 for instance).
Show me my Facebook wall faster than Gecko and Blink.
Show me my Twitter timeline faster than Gecko and Blink.
Show me the new Flickr faster.
Just to show off with journalists, show TechCrunch faster than Gecko and
Blink.
Consistently beat Google (Chrome) at showing a Google (search engine)
result page. (these are just examples of course).
These pages are pretty heavy in content and/or script and are the first
things that people will look at when using a browser written on top of
Servo.
If you start optimizing for these and keep as continuous integration
some version of these websites, you'll notice perf regressions on
real-world-like benchmarks.
As soon as perf becomes stable on these, expand to Alexa 100 (1000?).
Then fork the road to start exploring benchmarks that push Servo where
other browsers can go while stabilizing for the Alexa 100.
Also, a problem I see with perf benchmarks at this stage is that the
lack of feature and standard conformance might be biasing perfs a lot.
As an example, you don't de-optimize if the iframe and parent set
document.domain to the same value. I am not an expert in the topic, but
I can imagine that the support of CSS features or maybe even some
selectors makes big differences in perfs (I heard fixed background are
annoying to optimize for. Are they supported in Servo?), so a comparison
with current browsers may not be entirely accurate for now.
It'd be interesting to have an idea of how much features currently
missing make Servo performance better. I feel unable to judge how bad
the bias is, but it's there.
Focusing on Alexa 10-15, then Alexa 100 proves first a certain level of
standard compliance (IE8, I guess ?) and amount of available features
and the bias I mentioned starts fading away.
David
_______________________________________________
dev-servo mailing list
dev-servo@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-servo
