Friday, June 19, 2009

GWT / AJAX Performance Tools

Performance is important in your Web applications, especially in your client code. If you're a GWT user, I wanted to highlight a particularly useful benchmarking library - Jiffy and it's Firebug counterpart. It is the case that Safari's developer tools and Firebug by itself make user-injected benchmarking code less relevant for macro-benchmarks. But, if you're interested in fine-grained measurements and creating a rolling log of them, I found the developer tools a bit lacking.

Benchmarking is really just about specifying two points - a start and a stop. Giving that benchmark a label can be useful, and being able to review logs of these benchmarks over time would also be nice. Jiffy does all of this, with minimal setup time, and if you install the Firebug plugin, very immediate results.

First I downloaded the jiffy.js source and plugged this script tag into the HEAD of my GWT application's main HTML page:

script language="javascript" type="text/javascript" src="jiffy.js"

Then I created this simple class to wrap the required Jiffy calls as Java methods:

public class JiffyUtils {
public native static void mark(String markerName) /*-{
if( $wnd.Jiffy )

public native static void measure(String measurementName, String markerName) /*-{
if( $wnd.Jiffy )
$wnd.Jiffy.measure(measurementName, markerName);

...and I was ready to go! I inserted numerous marks and measures into a particularly slow, esoteric series of operations and determined where my bottlenecks were; oddly enough, it lead me to the discovery of the slowness of HashMaps in GWT 1.4 and, to a lesser extent, in 1.5 (Google for "HashMap optimizations gwt" for a bit more info.

Turns out I didn't need continual performance measurements, but Jiffy does offer a bunch of support for it. You can either get a JSON object out of Jiffy with all of your measurements by wrapping a call to
in a JSNI method, which I used momentarily but did not rely upon heavily. You can also use the Jiffy Batch and Realtime logging, which is explained on their site and would be useful for those interested in maintaining backlogs of their performance metrics for analysis.