[
http://jira.jboss.com/jira/browse/JBMESSAGING-1162?page=comments#action_1... ]
Gabriele Garuglieri commented on JBMESSAGING-1162:
--------------------------------------------------
Hi Tim,
perhaps i didn't explain myself well.
Depending on the number of messages persisted in db there's a minimum value for
FullSize under which JBM will refuse to start.
For instance from several runs we got the following results:
Messages min FullSize
180000 8000
300000 8188
600000 8880
900000 9500
1428188 26188
More or less linear until 900000 messages but then the line heads up. Even worst it
doesn't even have constant slope, with 1428188 msgs after a first good start with
26188 on the second run JBM refused to start unless FullSize was larger than 36000. (i
lost the log of that run and don't remember exact value, but it was slightly more than
36000).
Definitely there is a relation between minimum FullSize value and number of persisted
messages, but once the relation changed from linear to something else and i started
getting changing values for the same amount of messages it was impossible to determine
it.
Hence my request to document it.
And again max value for PageSize and DownCacheSize looks to have a relation with FullSize.
And IMO also this should be documented.
When you have to deal with very large number of messages these values are crucial for
optimal performances and deserves a better documentation.
My second point was that doesn't matter how well i tune Oracle, if i'm forced to
serially load in memory FullSize amount of messages for each queue/topic from a db that
contains tenth of millions of messages i'll always count startup times in tenth of
minutes.
Hence my second question, why JBM needs to serially load messages for each queue/topic
before startup instead of being able to parallel load messages at first queue/topic
access?
Thank you, Gabriele
JBM takes eons to startup with topic persisted messages within db
-----------------------------------------------------------------
Key: JBMESSAGING-1162
URL:
http://jira.jboss.com/jira/browse/JBMESSAGING-1162
Project: JBoss Messaging
Issue Type: Bug
Components: Messaging Core Persistence
Affects Versions: 1.4.0.SP1
Environment: Java 1.5.0_14
JBoss 4.2.2
Solaris 10, AMD Opteron biprocessor machine
Reporter: Gabriele Garuglieri
Assigned To: Tim Fox
Priority: Critical
The problem is reproduced as follows:
1-Start server
2-start a producer that inserts persisted messages (we tried 90.000 messages) in
TestTopic topic with default configuration where there are some consumers subscribed but
not active
3-stop producer and orderly shutdown the server
4-start server
During server startup JBM executes a select to load the persisted messages id's then
executes 180 selects loading 500 messages at time.
With 90.000 messages those 180 selects takes about 10 minutes. The DBMS is Oracle 10g
with dedicated instance on a separate machine with no more than 1% cpu load during test
and physical db on a very fast NAS.
While JBM is executing those selects, obviously the server startup hangs until the load
is finished.
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://jira.jboss.com/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira