Benchmark

To get an idea of how StAXON performs compared to alternatives, I did a simple performance test using the JUnitBenchmarks code benchmarking tool. The test compares reading and writing documents with

• XML StAX (implementation shipped with JDK 1.6)
• StAXON 0.5.0 (with Jackson 1.8.5 behind)
• Jettison 1.3 (another XML/JSON mapper)

Document

The test document corresponds to a perfect binary tree of height 15 (with 2¹⁶ - 1 nodes):

<alice xmlns="http://foo">
	<bob david="edgar>
		<bob david="edgar>
			...
		</bob>
		<bob david="charlie>
			...
		</bob>
	</bob>
	<bob david="charlie>
		<bob david="edgar>
			...
		</bob>
		<bob david="charlie>
			...
		</bob>
	</bob>
</alice>

or

{
	"alice" : {
		"@xmlns" : "http://foo",
		"bob" : [ {
			"@david" : "edgar",
			"bob" : [ {
				"@david" : "edgar",
				...
			}, {
				"@david" : "charlie",
				...
			} ]
		}, {
			"@david" : "charlie",
			"bob" : [ {
				"@david" : "edgar",
				...
			}, {
				"@david" : "charlie",
				...
			} ]
		} ]
	}
}

respectively.

Results

Some notes:

• The Write column refers to pure streaming mode, without caching events to determine array boundaries. This mode is not available with Jettison. With StAXON, array starts can still be triggered using the <? xml-multiple ... ?> processing instruction.

• The Write[] column refers to writing mode with automatically determining JSON array boundaries. As there's no such concept like arrays in XML, this is available only when writing to JSON. This is enabled in StAXON by setting the PROP_AUTO_ARRAY property on JsonXMLOutputFactory to true.

• Each test has been run 10 times: 5 rounds to warm-up and another 5 to take the average time from. Time spent with garbage collection is automatically excluded by the benchmark tool.

• The Jettison category was recorded using the Badgerfish mapping convention.

Source

View the Benchmark Test source.