Hi,
[Taking a private discussion to the list as others might be interested
as well.]
Edgar Poce wrote:
btw, I'm working on adding a commad and some jmeter functions for
traversing the tree. I plan to commit it tomorrow.
Very nice, I'll sure take a look at your code. I've just been setting up
a proper test harness and am currently planning the types of tests I
want to start running.
reading the entire tree example
/
/traverse [depth=-1] --> command
/while [__hasNextNode()] --> controller[function]
/ __nextNode() --> function
/ read node --> command
does it make sense? what other jmeter component do you think it would be
useful to extend in order to facilitate the benchmarking?
Yes, such a component would definitely be useful. Other nice components
would be a configurable query command and simple import/export commands.
I'm also about to write an alternative repository setup command that
contacts a remote repository using JCR-RMI.
I'd like to build a reasonably complete test script that would simulate
the access patterns of a typical web application. The script would
traverse the repository and perform some operations on each traversed
node. To cover the main access patterns the script should be
parametrized over the following variables:
1. Type of traversal. Either a full depth first tree traversal or a
configurable XPath query traversal.
2. Access mode. Read only (no changes to content), mostly read (update
only a single scalar property per node, e.g. an access timestamp), and
mostly write (modify all properties of the nodes).
Thus, to simulate a typical web site, I'd select some to traverse some
queried subset of a repository using either the read only or mostly read
access mode. For batch reads and writes I'd use full tree traversal with
the read only or mostly write access mode. I'm planning to run all these
tests both with and without the JCR-RMI layer to get a better feeling of
the RMI performance impact.
Please let me know if you have suggestions on how such test cases should
be implemented in JMeter.
I'm also trying to figure out a best example content for populating the
test repository. I'd be quite interested in trying to use at least parts
of the Wikipedia database dump for some *serious* stress testing.
Verifying whether Jackrabbit is capable of handling such amounts of
content would be an excellent scalability proof. ;-)
BR,
Jukka Zitting
