Performance

XML Manager is a data processing component so good scalable performance is an important design requirement. We use a randomised set of XML files, each 10 times larger than the first, to test the performance of the component.

The performance test is run on a AMD Athlon 2800 CPU, with the Java 1.4.2 runtime from Sun Microsystems. We obtained an average of over 5000 data records per second. Additional operations such as performing database INSERTS will of course reduce this figure.

Test Files

A typical row in each test XML file looks like this:

<row char="[">n^T4pmr2sbMn2Ew`@LA,br\Yz{U]<int>1384182710</int>
  <val long="-3591565520858899572">
    <double>0.34552274705395747</double>
    <date>1/31/11 2:27 AM</date>
  </val>
</row>

The following RecordSpec is used:

RecordSpec rs = new RecordSpec("/data/row", 
  new String[] { "@char", "int", "val/@long", 
                 "val/double", "val/date", "." },
  new String[] { "Char", "Int", "Long", 
                 "Double", "Date", "Text" } 
);

Files of size 10 to 1000000 records are created, and the five standard input methods are tested: List of String arrays, List of Lists, ResultSet, TableModel, and Java Beans. Each file is read 10 times and the timing results averaged.

Results

10 Records, time in milliseconds

Where the bars, from left to right, correspond to:

•   load method
•   loadAsLists method
•   loadTableModel method
•   loadResultSet method
•   loadBeans method

The remaining results are:

100 Records, time in milliseconds

1000 Records, time in milliseconds

10000 Records, time in seconds

100000 Records, time in seconds

1000000 Records, time in seconds

Analysis

First we divide the time taken by the number of records, to demonstrate that the data rate remains basically constant.

Second, we take log₁₀ of the times, to show that the parsing algorithm is 0(N).

Data

The original data used to produce these graphs is available as a CSV file so that you can verify the results for yourself. In addition, we provide the source code of the performance test class and the test file creator class so that you can run the tests to confirm these results.