PDA

View Full Version : IE Performance / Browser Benchmarks



MrSparks
26 Oct 2011, 1:23 PM
My Benchmark Update for 4.1 Pr1

Big improvement over the initial 4.0.0 release. (Well done Sencha Dev!)

IE8 : There still quite a way to go on IE8, thats currently 2.5 times slower under 4.1 pr1 than it was under 3.3.1

Chrome : Approximately 0.7 times slower under 4.1 pr1 than it was under 3.3.1

FF : Good news here, running approximately 0.7 times faster under 4.1 pr1 than it was under 3.3.1

Opera : Approximately 0.2 times slower under 4.1 pr1 than it was under 3.3.1

Here's a break down of the numbers. (image attached)

Benchmarks were measured against "Themes Example"

Best Mr Sparks

dongryphon
28 Oct 2011, 12:20 AM
I assume the numbers are in seconds?

The machines on which we are testing put Themes between 2 and 2.4 seconds for 4.1 and 2x+ that for 4.0.7. Curious what you are using hardware-wise. We did experiment at one point with VM's but found them unreliable for benchmark testing and have since purchased several real machines (one for each IE version)

MrSparks
28 Oct 2011, 12:53 AM
I assume the numbers are in seconds?

The machines on which we are testing put Themes between 2 and 2.4 seconds for 4.1 and 2x+ that for 4.0.7. Curious what you are using hardware-wise. We did experiment at one point with VM's but found them unreliable for benchmark testing and have since purchased several real machines (one for each IE version)


@ dongryphon

Figures are in seconds.

My system (Physical Machine) specs are:

Intel Pentium 4 - 3.2GHz
4GB Dual DDR - CL2.5
300GB 10K SATA - Raid 1
NVidia GeForce 7800 GS

Agree VM’s are a little unreliable if the host has other workloads outside of the benchmarks tests.

Happy run profile/test/provide as much info as you need.

MrSparks
28 Oct 2011, 1:11 AM
For comparisons sake, I've just ran 4.1 on what I would consider to be a high spec box. i.e. higher than what most business users would have access to.

2x Intel Xeon 2.5Ghz (Dual Core)
4GB DDR3 - CL3
NVIDIA Quadro VNS 290
300GB 10k SATA - Raid 1

4.1 Themes under IE 8 averages out at 4.25 Seconds
3.3.1 Themes under IE 8 averages out at 1.76 Seconds

Still seeing approximately 2.4 x speed reduction under 4.1

rich02818
28 Oct 2011, 4:45 AM
Don, on the same dev machines, what are the times for 3.3.4? Bettering that time is the actual target for business usefulness.

ZachG
28 Oct 2011, 3:46 PM
We did experiment at one point with VM's but found them unreliable for benchmark testing and have since purchased several real machines (one for each IE version)

How are they compared to what you see in IE Tester (http://www.my-debugbar.com/wiki/IETester/HomePage)? That's what we use and it's a pretty close approximation.

saprot
29 Oct 2011, 12:32 AM
+1


Don, on the same dev machines, what are the times for 3.3.4? Bettering that time is the actual target for business usefulness.

edspencer
29 Oct 2011, 6:28 PM
I think we can share that (not in the office right now so I don't have access) but be aware it's not quite apples for apples (there are more components in the 4.x version).

MrSparks
30 Oct 2011, 4:46 AM
I think we can share that (not in the office right now so I don't have access) but be aware it's not quite apples for apples (there are more components in the 4.x version).

I can't see any more components on the themes example under 4.1 pr1, so as another I've compared the most basic of the Sencha examples that could be classified an "app" on IE 8 and Chrome 15

IE 8
4.1 Complex layout under IE 8 averages out at 2.453 Seconds
3.3.1 Complex layout IE 8 averages out at 0.576 Seconds

Chrome 15
4.1 Complex layout under Chrome averages out at 0.965 Seconds
3.3.1 Complex layout Chrome averages out at 0.385 Seconds

Ok now an "ultra simple" comparison using the Sencha “Absolute Layout with Forms” example.

IE 8
4.1 Absolute Layout with Forms under IE 8 averages out at 1.52 Seconds
3.3.1 Absolute Layout with Forms IE 8 averages out at 0.45 Seconds

Chrome 15
4.1 Absolute Layout with Forms under Chrome averages out at 0.42 Seconds
3.3.1 Absolute Layout with Forms Chrome averages out at 0.333 Seconds

I think there something more "fundamental" in 4.1 that’s causing the performance issues. Apples for Apples, 4.1 pr1 (IE or Chrome) isn’t performing as well as it did under 3.3.1

MrSparks
30 Oct 2011, 5:33 AM
Possible cause if perf issues.

I've just been reviewing the Senchacon 2010 layout video. Jamie talks about an issue spotted in 3.x that was causing a huge overhead.

See 23:00 onwards
http://vimeo.com/17917111 (http://vimeo.com/17917111)

Specifically Jamie talks about the following erroneous function.


// private ==> used by Fx
adjustWidth : function(width) {
var me = this;
var isNum = (typeof width == "number");
if(isNum && me.autoBoxAdjust && !me.isBorderBox()){
width -= (me.getBorderWidth("lr") + me.getPadding("lr"));
}
return (isNum && width < 0) ? 0 : width;
}


Looking through the 4.1 pr1 release this function is commented out on debug with comments, however on debug and ext-all its not commented out. Could be nothing but thought it was worth a mention.

WixSL
31 Oct 2011, 6:02 PM
+1.

That's an interesting question.

Off-topic: Will it be Sencha on-line videos of this year's conference posted any time soon?

misyed01
1 Nov 2011, 8:49 AM
So we looped 25 array grids (from the ext examples) in an absolute layout using various ext versions. So here's how we tested the timeings:

1. Opened a new IE6 window with "about:blank" in it.
2. Loaded our local file path of the html file with the grids and ext version (all static files, no HTTP)
3. Captured the time in ms to load
4. Reloaded the page and kept capturing the time

Here is where we noticed a major memory leak in 4.1.pr1:

Load 1 - 4312ms
Reload - 5437ms
Reload - 6922ms
Reload - 8407ms
Reload - 10188ms
Reload - 11109ms
Reload - 12828ms
Reload - 14469ms
Reload - 16047ms
Reload - 17110ms

In Mozilla and Chrome, it was consistant, but IE6 was leaking real bad. Ext 4.0.2a and 4.0.7 remained consistent at about 10000ms each for the same set of tests.

Ext 3.4 averaged 1683ms to build the same thing, which is embarssing for the ext 4.x series. So even at best, Ext 3 was about 3 times faster than the tuned ext 4.1-pr1.

As we've migrated several of our corporate apps from Ext 3 to Ext 4.x, performance has been a MAJOR concern for us, across the board. I've been with you guys since version 1 and hope you guys can sort this out as it is very damaging to the reputation of the framework and product.

Please help!!!

NOTE - The column header lines (splits) were also not rendering...