Mike Pinkerton wrote:

> Just doing some _very_ cursory testing, it appears that the macEmbed 
> harness (a powerplant app that embeds mozilla) is faster at layout out 
> pages than mozilla navigator. Here's what I see with two equivalent 
> debug builds (same tree):
> 
> - first, I commented out the throbber in macEmbed, which is an animated 
> gif using ImageLib. In macEmbed, loading cnn.com locks the machine for 3 
> seconds (this is a debug build) but then the entire page displays in one 
> shot -- images and all. At this point it's done.
> 
> - with mozilla, it locks up for the same 3 seconds, but what displays is 
> just the webpage text, the images have yet to load. The images then 
> trickle in over the next 5 seconds.
> 
> - then i put the throbber back into macEmbed, and the images all loaded 
> more slowly (they weren't there at the initial blit of the page, but 
> finished loading more quickly).
> 
> Why are images loading so slowly? It seems that animated gifs of any 
> kind _really_ hurt our performance, especially page loads.

I ran debug versions of mozilla and macEmbed (PP Browser) built today 
from the same tree. Testing was done loading www.cnn.com. I wiped the 
cache folder of each program before running each test. I have a 1.5 
MBit/sec DSL connection so all of my times are a bit longer tan Mike's.

* With mozilla, most of the page came in in about 9 secs, then the 
images finished and the page fully loaded in 15 secs.

* With macEmbed, it did not display in one shot. It drew in the same way 
as mozilla but the times were 7 secs for most of the page and 10 secs 
for all.

   I then copied the sleep time calculation from nsMacMessagePump and 
put it in macEmbed. This didn't seem to make a big difference - only a 
second or so. This is where I thought the difference would lie, but 
apparently not.
   I then commented out the animated gif throbber in macEmbed, and that 
made very little difference.

So, our measurements are different but one thing is in common: The 
embedding app is faster. I am not sure XUL is the culprit. Needless to 
say, it will take some more measuring.

-Conrad


Reply via email to