openjpa-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Christiaan (JIRA)" <j...@apache.org>
Subject [jira] Created: (OPENJPA-474) Memory overhead by openjpa
Date Thu, 27 Dec 2007 09:40:43 GMT
Memory overhead by openjpa
--------------------------

                 Key: OPENJPA-474
                 URL: https://issues.apache.org/jira/browse/OPENJPA-474
             Project: OpenJPA
          Issue Type: Bug
         Environment: Windows XP, Kodo 4.1.4, Sun JDK 1.6.2, ms sql server 2005, JTDS driver
            Reporter: Christiaan


I am doing some analysis for the memory usage since we are running into a couple of problems
in our project when doing import of large quantities of objects. I've created a simple testcase
which does the following: 

1) Create all objects (2 classes, one class with 5 string attributes, the other is embedded
and has no attributes); 
2) Call makePersistent() on the objects; 
3) Call commit(); 

The results: 
- After step 1) takes up 23 mb of memory in the testcase; 
- After step 2) the memory usage has increased to 191 mb. Actually there are 3 scenarios I
used: 
1) Creating objects without the embedded class (which has no attributes, just an empty class):
75 mb 
2) Creating objects with the embedded class instantiated: 180 mb; 
3) Creating objects with the embedded class instantiated and after makePersistent() again
instantiated (overwriting the previous value): 191 mb 
- During step 3) commit() the memory usage peaks at 580mb 

So, 
1) to persist objects which take up 23mb of memory, a total of 580mb so giving a memory overhead
of a factor 25 by the openjpa! 
2) maintaining the datastructure whether objects are persistent increases the memory with
a factor of 4 to 8, depending whether embeddable objects are used; 
3) "overwriting" embedded objects within a transaction, the previous embedded object still
seems to take up some memory; 

I haven't done a thorough analysis on this but some interesting things I noticed from the
profiler: 
1) One of the hotspots is in storing the RowImpl._sql. If I am not mistaken, this sql is stored
(cached) for each object to be inserted. So in large transactions where a lot of similar objects
and modifications are involved, this is quite a duplication which leads to unnecessary memory
overhead. Of course there is a performance gain of caching it, but may be this should be better
balanced with memory usage?; 
2) Though I've set kodo.Log=none there is a hotspot in LoggingConnectionDecorator? Is this
decorator used for something else as well or is this a bug? I removed it in code from DecoratingDatasource
and this reduced the total memory usage to 500mb 

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message