STONE,ROBERT (HP-SanDiego,ex1) wrote:
I finally mastered latest build of xindice1.1b installed as a tomcat app on W2K platform. I understand that the "official" line is to use 1.1 rather than 1.0, however speed of access is very important for the application I'm developing. So far I'm seeing this when executing the same XPath query on the same collection:
1. Executing query on raw(non-indexed collection) 1.0 - 16500 ms, 1.1 - 63540 ms
2. After adding indexes - 1.0 - 400 ms, 1.1 - 1016 ms
As you can see I'm measuring time right before and after the query so I can blame the time increase on the slower startup time (can I?)
The collection consists of roughly 6500 simple xml files between 2 - 7K
each. Such performance degrading will almost definitely swing the pendulum towards
1.0 for my project
Can anyone comment on this?
These figures are somehow discouraging, and I'm wondering what could be the problem. Internally XIndice hasn't changed *that* much to justify such an speed loss, so I'd try a couple of paths in order to understand what's happening:
1. try the embed driver: this will give you access to the "raw" Xindice performance, without any transport overhead (it might well be XML-RPC serialization/deserialization, though I seriously doubt it);
2. check to see where most of the time is spent. I'd suspect something in Xalan code during the XPath search.
OTOH, please consider that, while being important, raw speed is not significant overall when it comes to scalability. Given the performance of the CORBA layer under heavy load, personally I'd me more than glad to trade some speed (agreed, not *that* much) for more scalability under load.
I will however do some tests myself and report back.
Ciao,
-- Gianugo Rabellino Pro-netics s.r.l. http://www.pro-netics.com
