Hi,
We used to use Xerces-c 3.1.4 before migrating to 3.2.2 recently. We found
3.2.2 is slower than 3.1.4 when processing a XML file with around 13,000
simple nodes.
On Windows, I wrote a performance testing program to test two different
ways of XML processing. One is based on DOMNode::getNextSibling(in function
cacheAll_getNextSibling), another one is based on
DOMNodeList::item(cacheAll_GetItem). There are two projects test these two
functions. Project TestXMLPerformance uses Xerces-c 3.2.2.
TestXMLPerformance.Xerces3.1 uses Xerces-c 3.1.4.
For cacheAll_GetItem, xerces 3.2.2 based project needs 8 seconds to finish,
however 3.1.4 based project only needs about 2.5.
DOMNodeList::item consume CPU time very high in both 3.1.4 and 3.2.2. But
the newer version is worse. Large part of CPU time are consumed by memory
allocation and copy in _RTDynamicCast. In the stack trace, AnsXML is the
wrapper of Xerces and Xalan.
My questions:
1. Is 3.2.2 supposed to be slower than 3.1.4, because it has more features
and security enhancement?
2. Are the switches used in the configuration and build made new library
slower. 3.1.4 have .sln & .vcxproj came with the source code. So we used
Visual Studio to build the library. For 3.2.2, there were no Visual
Studio solution or project files provided. So we used
cmake -G "Visual Studio 15 2017 Win64"
-DCMAKE_INSTALL_PREFIX={PathToXercesInstall} -Dxmlch-type=wchar_t -S
{PathToXercesSource} to configure the project.
Then used
cmake --build . --config Release
ctest -V -C Release -j 4
cmake --build . --config Release --target install
to build and install.
Is it possible -Dxmlch-type=wchar_t made the library slow down.
3. Is it possible to improve the performance of 3.2.2 through different
build switches or options?
Thanks!
--
Qihong Wang
Senior Software Developer
ANSYS, Inc. www.ansys.com
Southpointe
2600 ANSYS Drive
Canonsburg, PA 15317-9565
Phone:724-820-3966
Fax:724-820-3816
E-Mail: qihong.w...@ansys.com
