SiTE - Performance Comparison
Environment
Test target was the "ULose" game (in the distribution under
demos/de/tivano/site/servlets ), running
in different configurations. The servlet was accessed through the
Enhydra multiserver version 3.1.1b1 running
on SUN jdk1.3.1. To run
the test we used the MakeHttpLoad tool with 10,000 requests made to the
starting page of the game and 10 simultaneous threads. (MakeHttpLoad is
included in the distribution under demos/de/tivano/tools .)
The tool was running on the same single-processor machine as the
Enhydra multiserver.
Measurements
Configuration | Time (s) |
Errors |
Cache | Parser | MetaRepository |
- | SimpleText | - | 54.421 | 0 |
ProbabilisticCache | SimpleText | - | 29.497 | 0 |
FilesystemPersistentCache | SimpleText | - | 92.845 | 0 |
Other technologies
Of course we're interested how our engine compares to others. Especially to
XMLC, which has given us much inspiration
on how to build a template engine (and also what to avoid).
Naturally, there are small differences in the servlet implementations for each
engine, as well as small differences in the templates. We have tried to model
the behaviour of them as closely as possible. See the "Remarks" column for
details. For the Enhydra and Tomcat tests we used the same JDK 1.3.1 as for
the above measurements. In our setup we had to use the "classic" VM for the
tests with Tomcat, while
for the tests with Enhydra multiserver we used the default "client" VM.
Environment | Template-Engine | Time (s) | Remarks |
Enhydra Multiserver | XMLC | 87.211 |
This can best be compared to the "ProbabilisticCache" setup measured
with SiTE. |
Enhydra Multiserver | SiTE-Servlet | 29.497 |
The configuration with "ProbabilisticCache" was used here. |
|
Jakarta Tomcat 4.0.1 | JSP | 194.516 |
The "classic" VM had to be used here. |
Jakarta Tomcat 4.0.1 | SiTE-Servlet | 118.980 |
The configuration with "ProbabilisticCache" was used here. |
|
WebSphere 3.5 | JSP | 47.276 |
IBM JDK 1.2.2 |
WebSphere 3.5 | SiTE-Servlet | 46.071 |
IBM JDK 1.2.2 |
Since it's pretty pointless to measure recompilation times for JSP and XMLC,
we have decided to not include these values. They are so slow that developers
are well-advised to make sure compilation takes place before anyone
accesses the page.
[ t]ivano software gmbh
www.tivano.de
Enhydra XMLC is a registered trademark of Lutris Technologies, Inc. All other
products and services are trademarks of their respective owners.
|