Browsing speed improvement

Pascal · Post by **Pascal** » November 8th, 2002, 12:22 pm

According to the HTTP specs, only a limited number of simultaneous connections per web server are allowed. As usual, MS IE has an undocumented feature that allows you to - experimentally - increase that number. To do so - at least in Win2K and XP -, you must add a few entries to the registry:

HKEY_USERS\.DEFAULT\Software\Microsoft\Windows\CurrentVersion\Internet Settings\
"MaxConnectionsPerServer"=dword:00000020
"MaxConnectionsPer1_0Server"=dword:00000040

HKEY_CURRENT_USER\Software\Microsoft\Windows\CurrentVersion\Internet Settings\
"MaxConnectionsPerServer"=dword:00000020
"MaxConnectionsPer1_0Server"=dword:00000040

In theory, and although those values work fine in most cases, they exceed the HTTP specs and therefore might cause problems with some websites. Now I must say that I've never experienced such problems and that, thanks to this hack, MS IE allows me to browse much faster than Phoenix.

Question: Isn't it something that might be added to Phoenix in the future. Obviously, not via a reg key, but that's a detail. Any ideas? Opinions ?

mboullet · Post by **mboullet** » November 8th, 2002, 12:36 pm

See Mozilla bug: http://bugzilla.mozilla.org/show_bug.cgi?id=101535
marked WON'T FIX.

If I remember well, Mozilla used to have more possible connections (in the old times: M8) and
benchmarks proved that they should reduce it to get better performance.

laszlo · Post by **laszlo** » November 8th, 2002, 12:37 pm

Just take a look at <Phoenix dir>/defaults/pref/all.js. Look for the prefs starting with "network.http.max-". If you'd like to change one of those, prepend "user_" to them, add them to user.js in your profile dir (create the file if it doesn't exist) and restart Phoenix. You can also change prefs.js without Phoenix running.

I doubt that it will actually make things "much faster" though...

Stefan · Post by **Stefan** » November 8th, 2002, 12:42 pm

Well, breaking the specs sounds a little troublesome.
It only works as long as not everybody does it (then I assume it would be much slower for everybody).

In any case, would be nice to get some figures on exactly homw much of an improvment this adds. Try and test eg IE- default, IE hacked, Phoenix and K-Meleon (which is the fastest Win Gecko browser).

http://www.numion.com/YourSpeed/

Pascal · Post by **Pascal** » November 8th, 2002, 2:05 pm

Stefan wrote:Well, breaking the specs sounds a little troublesome.

Agree. It's the choice that MS made for most - if not all - of its internet-related products. Understand that nobody likes it. It's the good old tradeoff between standards and performance. Ever wondered why they're leading the market?

Stefan wrote:It only works as long as not everybody does it (then I assume it would be much slower for everybody).

Well, if everybody does it, then some of them may end up having to wait a bit for the web server to be available. But this hack has been around for a while - at least since the advent of Win2K (1999 ?) -, so either I'm "the only lucky one" (which I sincerely doubt of) or...

Stefan wrote:In any case, would be nice to get some figures on exactly homw much of an improvment this adds. Try and test eg IE- default, IE hacked, Phoenix and K-Meleon (which is the fastest Win Gecko browser).

I tried it already - hence, this thread - with IE default, IE hacked and Phoenix default. Obviously I don't have a chronometer in my hand, but it "feels" much faster with IE-hacked than with any other solution. I invite all W2K or WinXP users to test it and get back here to give us their opinion. Now, of course, one can always decide to stick to standards and religiously hope that one day Microsoft will loose - or even abandon (ha ha ha) - a bit of its 96% worldwide share of the Internet browser market ...

Stefan · Post by **Stefan** » November 8th, 2002, 2:37 pm

Pascal wrote:Ever wondered why they're leading the market?

No, I'm fully aware of that MS used their monopolistic position in the OS scene to kill of all opposition. Luckily they didn't succede fully.

Stefan wrote:It only works as long as not everybody does it (then I assume it would be much slower for everybody).

Well, if everybody does it, then some of them may end up having to wait a bit for the web server to be available. But this hack has been around for a while - at least since the advent of Win2K (1999 ?) -, so either I'm "the only lucky one" (which I sincerely doubt of) or...

Well, it's not too common for people to hack their registry. If it becomes a default option in a browser most users will will start turn it on if makes the browser perform better.

Stefan wrote:In any case, would be nice to get some figures on exactly homw much of an improvment this adds. Try and test eg IE- default, IE hacked, Phoenix and K-Meleon (which is the fastest Win Gecko browser).

I tried it already - hence, this thread - with IE default, IE hacked and Phoenix default. Obviously I don't have a chronometer in my hand, but it "feels" much faster with IE-hacked than with any other solution.

That webpage have a "chronometer" built in

When I tested K-Mel blew everything else away.
Didn't try with a hacked IE though.

Of cource no test is the ultimate once and for all test of preformance, but AFAICT thatone does at least reflect my personal experiences in hard numbers.

laszlo · Post by **laszlo** » November 8th, 2002, 2:46 pm

Pascal wrote:[...] Obviously I don't have a chronometer in my hand, but it "feels" much faster with IE-hacked than with any other solution. [...]

That's nice, but pretty worthless, objectively. I also tried this way back when I used IE more often, but I was never able to find any measurable speed gains. That's why I already expressed my doubts about this.

BTW: If you want to try a performance improvement that's way smarter and included in the HTTP/1.1 spec (see 8.1.2.2):

user_pref("network.http.pipelining", true);
user_pref("network.http.proxy.pipelining", true);
user_pref("network.http.pipelining.maxrequests", 8);

That's what gives me a little speed boost here. It's something that's not even implemented in IE.

Arvid Axelsson · Post by **Arvid Axelsson** » November 8th, 2002, 4:25 pm

user_pref("nglayout.initialpaint.delay", 0);
Makes Phoenix/Mozilla feel faster to me.

laszlo · Post by **laszlo** » November 8th, 2002, 4:53 pm

Arvid Axelsson wrote:user_pref("nglayout.initialpaint.delay", 0);
Makes Phoenix/Mozilla feel faster to me.

I also browse with this setting. This is comparable to increasing the number of simultaneous connection: it effectively raises the total page load time, because more reflows are needed, but it feels faster, because the browser starts painting straight away.

oceanwaterz · Post by **oceanwaterz** » November 9th, 2002, 11:37 am

Do I have to add this to my prefs, or should it be already there ?

I can't seem to find it if it is....

oceanwaterz · Post by **oceanwaterz** » November 9th, 2002, 11:45 am

Found it !!

worth · Post by **worth** » November 9th, 2002, 12:19 pm

In which file does that have to added?

oceanwaterz · Post by **oceanwaterz** » November 9th, 2002, 12:33 pm

As "laszlo" mentioned earlier...

Phoenix-defaults-prefs-all.js..

Then add user_ before pref

Stefan · Post by **Stefan** » November 9th, 2002, 2:30 pm

worth wrote:In which file does that have to added?

http://texturizer.net/phoenix/edit.html

laszlo · Post by **laszlo** » November 9th, 2002, 8:57 pm

oceanwaterz wrote:As "laszlo" mentioned earlier...

Phoenix-defaults-prefs-all.js..

Then add user_ before pref

Sorry if my post wasn't clear about this. I didn't suggest to add it to all.js (which is where Phoenix gets its defaults for all profiles from). This file is useful to get some informations about prefs.

Please add it to user.js or prefs.js in your profile dir. The link Stefan provided will guide you through this.