Comments on: Optimizing Page Load Time (and a little about the Debug menu)

By: cyrilgodefroy

cyrilgodefroy — Wed, 04 Jul 2007 21:23:24 +0000

For those reading this post after June 2007, WebKit now has a wonderful Web Site Inspector. It’s available on Safari for Mac or PC. And apparently Safari now parallelizes the download of css files, maybe even js. Maybe another reason it’s faster than some other browsers?

By: mdknapp

mdknapp — Mon, 11 Jun 2007 20:22:00 +0000

To enable the debug menu in Safari 3 for Windows:
===============================
In the file:
C:\Documents and Settings\USERNAME\Application Data\Apple Computer\Safari\Preferences

Add the following:
IncludeDebugMenu

By: aaron42net

aaron42net — Mon, 22 Jan 2007 06:13:33 +0000

One interesting quirk of many modern browsers (including at least Firefox/2 and Safari/419.3) is that external javascript and CSS is fetched serially rather than making use of the 2 or more parallel connections available.

As an example, Slashdot seems to load 5 external stylesheets and 12 external javascript URLs, while Digg loads 2 external stylesheets and 21 external javascript URLs. Loading these serially means at least 17 and 23 round-trips respectively between browser and server. At 100ms per round trip, this adds 1.7 or 2.3 seconds above any transfer times for the data. If Safari could spread these loads over 2 connections instead, these extra delays would be cut in half.

I understand that both CSS and Javascript have to be applied/executed in a specific order. But how hard would it be to begin fetching them sooner?

— Aaron

By: heckenpenner_rot

heckenpenner_rot — Sat, 04 Nov 2006 10:58:25 +0000

How do you guys manage connections? Is there some sort of connection pool that keeps existing connections alive? Do you open multiple connections to the same hostname?

By: maciej

maciej — Thu, 02 Nov 2006 09:00:25 +0000

The problem is that you can’t tell when you are getting a bad response. The symptoms are things like a connection hang, or the results from multiple responses getting mixed together so you’ll get headers and data from more than one response mixed together. If it was possible to identify a bad response definitively we could do a retry thing.

By: Thinine

Thinine — Wed, 01 Nov 2006 01:20:59 +0000

Can’t you deal with bad responses when using HTTP pipelining? Just rerequest the page without using pipelining. Sure, slower for that page, but you get the speedup with all the rest. Or at least give us the ability to turn it on in WebKit/Safari.

By: Mark Rowe

Mark Rowe — Tue, 31 Oct 2006 18:47:06 +0000

xenon, downloading the site locally defeats the purpose in testing page load times over the network, don’t you think? 🙂

By: xenon

xenon — Tue, 31 Oct 2006 17:49:30 +0000

@sulka, the network will cause wide variance with the PLT. I recommend you download the site locally and disable the network and you will see very consistent times. Other things like Spotlight should be disabled to prevent outside slowdowns also.

By: ahruman

ahruman — Tue, 31 Oct 2006 12:33:19 +0000

I got weird values too… but it did tell me Safari was leaking memory. 🙂

By: maciej

maciej — Tue, 31 Oct 2006 08:58:04 +0000

The nodes and memory amount refer to changes in memory being used by the browser. They don’t relate to DOM nodes or amount of data downloaded. Sorry if that’s confusing!