One of the big questions that people have had about ExtCore is why didn’t it make use of the Sizzle selector engine. That was one of the questions I wanted to get answered at the Ext Conference last week, and I managed to pin down one of the core developers to get an answer. Their opinion was that Sizzle is too large and too slow. I don’t really care about the size aspect. They all tend to pack down into the 25K range, but I was interested in the speed comment. Like a lot of folks, I was led to believe that Sizzle had a lot of, well, sizzle. So in the interest of getting to the bottom of the story, I did a little unscientific testing of my own using Slickspeed, which tests selector engine performance.
I grabbed the latest of the main libraries: ExtCore (beta1), JQuery 1.3.2, Dojo Core 1.3.0, Prototype 184.108.40.206 and the Sizzle engine by itself. I ran Sizzle by itself because although JQuery uses Sizzle, the version of Sizzle in JQuery 1.3.2 is 0.93 while the current version on the Sizzle website is version 1.0. I used the latest version of Slickspeed and made three runs in IE8, Firefox 3.0.8 and Chrome 220.127.116.11. Yes, I disabled Firebug in Firefox. My dev desktop is a rather beefy Windows Vista 64-bit machine with a quad-core 2.66ghz CPU and 8GB of RAM. I did a complete page refresh between runs to start relatively clean. I was going to relaunch the browser between runs, but it ended up not making much of a difference.
First up was the latest and greatest from Microsoft, Internet Explorer 8.
- Prototype: 440ms
- Dojo: 61ms
- JQuery: 48ms
- ExtCore: 80ms
- Sizzle: 46ms
The image shows the results from the last run. I was actually surprised that IE8 did reasonably well. One thing is clear, friends don’t let friends use Prototype with Internet Explorer.
Firefox was a slightly different story. The selector engine in Prototype still got its clock cleaned, and ExtCore swapped around with JQuery and Sizzle:
- Prototype: 145ms
- Dojo: 64ms
- JQuery: 77ms
- ExtCore: 49ms
- Sizzle: 71ms
It was interesting both Sizzle implementations slow down in Firefox. I would wager that the Sizzle folks made a conscious decision to optimize the core selector engine for the most widely used browser (Internet Explorer). ExtCore leaps ahead with Firefox 3, while Dojo is still pretty consistently finishing in the middle. Oh, and friends don’t let friends use Prototype with Firefox either.
Chrome is an absolutely game-changing stunner of a browser. Based on the selector benchmarks, the developers for all these engines should be offering to have children for the Chrome developers:
- Prototype: 13ms
- Dojo: 7ms
- JQuery: 8ms
- ExtCore: 13ms
- Sizzle: 8ms
Even lowly Prototype with Chrome makes any of the other libraries running in Firefox look like they’re standing still. Chrome is simply ungodly fast with all them. ExtCore could probably still use some tweaking, but we’re talking about a 7ms delta between the fastest and slowest.
So the Ext team’s claims for Sizzle being slower are only partially true. Sizzle appears to be faster in Internet Explorer, while the story flips around in Firefox where ExtCore is faster. All of them, even Prototype, are fast in Chrome. Pray for the day when Chrome takes over the world. Until that time, It looks like all them (Dojo, JQuery/Sizzle, ExtCore) are good in IE and Firefox, with the nod going to JQuery/Sizzle for Internet Explorer, the browser market share leader. And you really, really don’t want to use Prototype until all your clients have moved to a modern browser with native selector engines, or Chrome.
One glaring omission is that I didn’t test with IE7 or IE6. Using either of those browsers is also on my “friends don’t let friends…” list. It would be safe to say that they would both be slower. As I don’t have either installed on any of my computers, I would be happy to add an update for anyone who wants to run the tests themselves.
The Ext team earns some slack in their Internet Explorer performance as they are only a beta compared to the others. I’ll rerun the test again when ExtCore goes GA. And on a final note, here is the config.ini file for Slickspeed in case anyone wants to try the benchmark themselves, or you can grab the whole Slickspeed install with the selector engines configured.