While playing with browsers last time on character encoding issues, I timed every browser loading 10000 non-existent image tags. As you may have guessed, the script is this :
<img src=non-existent&onerror=alert(1) />
Notice that the javascript is rarely involved unless the exploit succeeded. So for almost all the cases only the image tag is involved, and not the Javascript Engine.
The test is conducted on a VMWare Windows XP Single Core 2 GHz CPU with 256 RAM. A poor specification but nonetheless helps to magnify the problems in within.
With 10000 image tags on a single page, the test is conducted without closing the browser process for 6 iterations so as to capture possible memory leaks and consistent results. The results is as follows :
| Browser name | Time (seconds) | |||||
| Opera 9 | 61.2 | 90.9 | 44.2 | 54.5 | 47.2 | 58.2 |
| Chrome 0.2 | 102.6 | 120.3 | 109.6 | 122.9 | N/A | N/A |
| Firefox 3 | 41.8 | 50.1 | 45.1 | 52.1 | 57.4 | 66.6 |
| Internet Explorer 7 | 9.2 | 8.3 | 8.8 | 8.0 | 8.5 | 8.9 |
| Safari 3 | 84.2 | 79.0 | 138.4 | 352.6 | (>900)* | N/A |
* – Safari 3 only executed 3/4 of the scripts after 900 seconds, and was terminated because it is too slow.
Opera 9 – It scores consistently at about 50 seconds except for the 2nd iteration, which is quite good for 10000 tags. At such a poor specification, it is very responsive although it is heavily loaded.
Chrome 0.2 – It is slow at some 110 seconds, and scores consistently bad. The browser also literally hung and remains unusable during the load. As it is a beta, I am sure they will improve in time.
Firefox 3 – Fast at 50 seconds, but seems to have a slowing trend, could be caused by memory problems although formal memory profiling suggests otherwise [1]. Facts are facts. Overall experience is very responsive despite the heavy load.
Internet Explorer 7 – 8 seconds. Internet Explorer 7 seems to perform very well on such failing image tags, scoring consistently over the 6 iterations, and no signs of memory leakage. Probably optimizations are behind the scenes.
Safari 3 – An obvious growing trend from 80 seconds to exceeding 300 seconds, and took 900 seconds just to execute about 7500 of the image tags in the page at the 5th iteration. Obvious memory leakage and other problems are behind. The browser stays responsive, but its crawling speed is making it unusable.
Conclusion :
It seems to me the worst browser seems to be Safari here. If a formal performance test is conducted on such browsers, it may tell a lot more. But meanwhile, I do not have a reason to use Safari unless I want to test any new Carpet Bombing. =) Firefox 3 is still the road to go! And in China, inevitably the IE7 as well.
This is not a formal performance test and the test case consists of just a single type of script. I am just magnifying the case when a page has more images and seeing what are the capabilities of the browser is. I think IE7 should really get rid of that odd loading at the start of the browser, and maybe it can help pull back the users from Firefox 3.
—
References :
[1] http://blog.pavlov.net/2008/03/11/firefox-3-memory-usage/
0 Responses to “My Performance Test on Browsers”