Browsing speed improvement

Pascal · Post by **Pascal** » November 10th, 2002, 2:59 am

All,

Thanks for your helpful - at least for some of them - answers.

Regards,

Pascal

hiTCH-HiKER · Post by **hiTCH-HiKER** » November 11th, 2002, 9:31 am

That webpage have a "chronometer" built in
When I tested K-Mel blew everything else away.
Didn't try with a hacked IE though.

where is the URL?

Stefan · Post by **Stefan** » November 11th, 2002, 10:23 am

hiTCH-HiKER wrote:
That webpage have a "chronometer" built in
When I tested K-Mel blew everything else away.
Didn't try with a hacked IE though.

where is the URL?

4th post from the top.

Pascal · Post by **Pascal** » November 11th, 2002, 12:01 pm

Hi Stefan,

I just tried your speed measuring web site (as a reminder to everyone, it's http://www.numion.com/YourSpeed) and did the following test:

Testing conditions: Pentium III 450 MHz with 384M, with W2K SP3 and a 512kbps ADSL link
Loaded both Phoenix and MSIE 6 SP1 in memory, activated a test ("from France") with first Phoenix (build 20021108), waited for results, then IE and waited for results, then tried again with Phoenix to recheck the results.

Now here's what I got:
* Phoenix (no hack): throughput 16775 Bps or 134 kbps
* MS IE (hacked the way I said in my initial post): Throughput 36078 Bps or 289 kbps

That's more than twice Phoenix's speed.

So Stefan, here's my question, to you and to any reader on this forum -- I have no idea how technically serious these tests are (i.e. from the website's standpoint) and I totally agree with you on the fact that this hack breaks the specs, and that breaking specs is no good, etc., but don't these results make you stop and think? Should Phoenix remain far behind MS IE in terms of browsing speed only in the Holy Name of the Specs?

It's an open debate - I hope - and would I suggest other people, from their current locations, to do the same comparative tests, both with and without the MS IE hack, just to see if they get a less impressive difference. But - and thanks to the site you advised us on, Stefan - for the moment we're not talking of just a 15 or a 25% speed increase here. It's a tremendous difference.

- Pascal

Ploum · Post by **Ploum** » November 11th, 2002, 12:08 pm

I don't understand !

I create the file user.js into de phoenix/defaults/profile folder, and it doesn't care about it ! What must I do ?

My user.js is :

// Stop reusing active windows:
user_pref("advanced.system.supportDDEExec", false);
// Click on throbber to go to Phoenix Help:
user_pref("browser.throbber.url","http://texturizer.net/phoenix/");
user_pref("network.http.pipelining", true);
user_pref("network.http.proxy.pipelining", true);
user_pref("network.http.pipelining.maxrequests", 8);
user_pref("nglayout.initialpaint.delay", 0);

Must I rename it prefs.js ? must I deplace it ?

Stefan · Post by **Stefan** » November 11th, 2002, 3:28 pm

Pascal wrote: but don't these results make you stop and think?

Definitly. That is too much to be ignored if it's not just a fluke on your setup.

Do you btw have code to do the hack for IE on Win98SE or a link to it?
Would definitly be interested in trying it out myself.

BTW, it might be good if you could also test IE unhacked just to give some indication of how myuch a difference it does.

laszlo · Post by **laszlo** » November 11th, 2002, 4:16 pm

Pascal: You should definitely provide values for IE with default settings... otherwise it's comparing apples to oranges.

You should also test Phoenix with the following settings (or similar):

user_pref("network.http.max-connections", 60);
user_pref("network.http.max-connections-per-server", 40);
user_pref("network.http.max-persistent-connections-per-server", 20);
user_pref("network.http.max-persistent-connections-per-proxy", 20);
user_pref("network.http.pipelining", true);
user_pref("network.http.proxy.pipelining", true);
user_pref("network.http.pipelining.maxrequests", 8);

shimage · Post by **shimage** » November 11th, 2002, 4:20 pm

i dunno what the hack for ie is. i'm using win2k SP3 on this machine, if maybe the hack is already in it already.

IE --> 302kbps
phoenix(1111) --> 184kbps

it's not a factor of 2 for me, but it's still quite noticeable.

@laszlo
i thought the point was that he thinks the hack should be used in phoenix too. wouldn't it be pretty useless to not compare phoenix to "hacked IE" in that case?

laszlo · Post by **laszlo** » November 11th, 2002, 4:24 pm

Ploum wrote:I don't understand !

I create the file user.js into de phoenix/defaults/profile folder, and it doesn't care about it ! What must I do ?

I assume that you've checked with about:config that the settings didn't make it through.

Method 2: Exit Phoenix (completely; this means that if you're using a Quick Start Feature hack, disable it; check with Task Manager in Windows or with ps or whatever you want in Linux). Copy the lines from user.js to prefs.js. Restart Phoenix. Check about:config. This should do it.

Ted Mielczarek · Post by **Ted Mielczarek** » November 11th, 2002, 4:25 pm

shimage wrote:i dunno what the hack for ie is. i'm using win2k SP3 on this machine, if maybe the hack is already in it already.

IE --> 302kbps
phoenix(1111) --> 184kbps

it's not a factor of 2 for me, but it's still quite noticeable.

@laszlo
i thought the point was that he thinks the hack should be used in phoenix too. wouldn't it be pretty useless to not compare phoenix to "hacked IE" in that case?

No, because there might be some other difference in IE's network code that is causing the noted difference, and the hack might not be significantly changing the results.

shimage · Post by **shimage** » November 11th, 2002, 4:34 pm

granted. that's why you test both (hacked/unhacked) and look at the difference.

that number was for without the first four lines of laszlo's suggested additions. i didn't keep track of the numbers, but when i repeated the test it appeared that the 184 was a bit high, actually.

i then added those four lines and got 244+/-47 (n=4), though there was significant drift (duh), so i don't think that sigma is even meaningful here. IE (presumably unhacked) was consitently right around 300.

laszlo · Post by **laszlo** » November 11th, 2002, 5:02 pm

shimage wrote:granted. that's why you test both (hacked/unhacked) and look at the difference.

Yes. The values you got seem to show that IE with default settings (which are a lot lower than with the registry change) already comes out much faster than Phoenix in this test.

shimage wrote:that number was for without the first four lines of laszlo's suggested additions. i didn't keep track of the numbers, but when i repeated the test it appeared that the 184 was a bit high, actually.

i then added those four lines and got 244+/-47 (n=4), though there was significant drift (duh), so i don't think that sigma is even meaningful here. IE (presumably unhacked) was consitently right around 300.

The last three (HTTP pipelining) won't do anything here. The test loads a page that fetches 40 images, each from a different server. There are not many comparable pages in the real world.

shimage · Post by **shimage** » November 11th, 2002, 5:35 pm

laszlo wrote: Yes. The values you got seem to show that IE with default settings (which are a lot lower than with the registry change) already comes out much faster than Phoenix in this test.

agreed; it's something else that's making IE so much faster. i also think that enabling the max-connection thingers do help somewhat, though. at least, more than you previously implied.

laszlo · Post by **laszlo** » November 11th, 2002, 7:11 pm

shimage wrote:[...] it's something else that's making IE so much faster. i also think that enabling the max-connection thingers do help somewhat, though. at least, more than you previously implied.

Not really. There's a severe error in reasoning here. As I said, this test loads 40 images, each from a different server. Thus raising the allowed number of connections per server (as the registry change for IE and three of the four prefs you changed for Phoenix do), won't help anything. In other words: the use of this test page to test the efficiency of the IE registry change is highly questionable.

The fourth pref changed the total number of allowed HTTP connections, which has a default of 24 in Phoenix. To clarify: an unchanged Phoenix would start to fetch 24 images in parallel, start to load the 25th when the first image is loaded and so on. The change allowed the browser to fetch all 40 in parallel from the start, which may be an explanation for the difference after the pref change.

IE might just have a higher default value here, which is perfectly allowed by the HTTP spec (it only deals with the number of connections to one server). This doesn't explain the size of the gap between a changed Phoenix and an unchanged IE though; that's something to talk about.

Ted Mielczarek · Post by **Ted Mielczarek** » November 11th, 2002, 7:35 pm

So essentially this test is completely irrelevant to the original topic, which was max connections to a server.

A test that tested said topic would be nice, although the performance discrepancy already noted would be interesting to look into as well.

Browsing speed improvement

Reply to all

huh...

Re: huh...

Testing