Improve IO on Paging / Depaging

Hi Clebert,

could you take look to the following problem, please?
I hit it during the testing which was related to JBoss Messaging. I wanted to compare performance JBM vs HornetQ with Specjms. I took existing test for HornetQ and I cloned it for another lab. I have changed hosts and initial load nothing else.

Test tries to execute specjms test and if execution is successful it tries to increase load. Test runs fine for several iterations but when the load is quite high test hangs up when undeploying/deploying queues for test between iterations (which was the first surprise). HornetQ server does not respond to test client and still produces logs [1]. Test client does not produce load at this time (load is almost 0) but server is overladed (2x dual core CPU, 390%).

Original test from Messaging lab is fine, no errors. Cloned test with modified host names and decreased initial load fails. I have executed it for several times and I hit this issue every time.

Messaging lab:
RHEL 6 + etx4 + faster CPU + AIO

Perf lab
RHEL 5.5 + etx3 + NIO

HQ Distribution:
EAP 5.1.1 ER1 with NIO.

[1]
2011-05-05 07:27:33,878 WARN [org.hornetq.core.paging.cursor.impl.PageSubscriptionImpl] (Old I/O server worker (parentId: 1658217180, [id: 0x62d662dc, perf01/10.16.88.179:5445])) Timeout on waiting cursor org.hornetq.core.paging.cursor.impl.PageSubscriptionImpl@7e95e4d6 to be closed
2011-05-05 07:27:33,878 WARN [org.hornetq.core.paging.cursor.impl.PageSubscriptionImpl] (Old I/O server worker (parentId: 1658217180, [id: 0x62d662dc, perf01/10.16.88.179:5445])) Timeout on waiting cursor org.hornetq.core.paging.cursor.impl.PageSubscriptionImpl@52605000 to be closed