Hi,
After some tweaks and fixes [1] I ran some benchmarks regarding
observation throughput with different Microkernels.
The test scenario adds roughly 60000 nodes in a background thread while
counting the number of observation events received. That number should
be twice the number of added nodes (one event for the node and another
for jcr:primaryType).
When charting the number of added nodes together with the number of
events received against time [2], it looks like SegmentMk and H2 can
keep up while with MongoMk there is a backlog of events developing: in
the former cases the red line has approximately the same slope as the
blue one. In the latter case both slopes are the same until all nodes
are in place and only the red line continues to rise as event delivery
catches up.
I'll commit the benchmark code later so we can re-run when necessary.
Michael
[1] http://svn.apache.org/r1494931
[2]
https://docs.google.com/spreadsheet/pub?key=0AqWOLyLNdfKedFFreEI3OGRfcy0waXFKbHhnZi12Snc&output=html
