MQSeries.net :: View topic

griffel

I have a small test program that uses a message selector to get
every other message out of a queue filled with 1000 messages -
that is it gets 500 messages in a single transaction.

The test (Java impl, JMS 1.1, WebSphere MQ 5.3 CSD10 for Windows)
does its job using 3 different strategies, measuring total run times for
each:

1. it uses a loop do to 500 Conumser.receive calls within a transacted session.

2. it uses a ConnectionConsumer (IBM's ASF class) with a dummy
MessageListener (noop onMessage) and an IBM-style ServerSessionPool
with a single session.
Batch size for the consumer is set to 1, that's it gets exactly 1 message
per session.

3. same consumer as above, but batch size set to 500.

I ran this test against 2 queue manager installations:
- MQ 5.3.1, all PTFs before 05/17/2005, zOS 1.6
- MQ 5.3 CSD 10, Windows Server 2003

Here are the results (milli secs):

Windows:
--------------------------------------------------------------
Receive [ms]: 5498
Consumer (single) [ms]: 4116
Consumer (batched) [ms]: 741

z/OS
--------------------------------------------------------------
Receive [ms]: 5128
Consumer (single) [ms]: 14701
Consumer (batched) [ms]: 17865

The windows scenario behaves as I've expected (and relied on).
The z/OS scenario is inacceptable.

Can anybody comment on this?