Performance between MQ and MQ JMS

Question

Hi! I now MqSeries is not the subject of this newsgroup, but, as soon as,
I'm using Weblogic to access MQ JMS queue, I'd like to now the opinion of
other people about performance in the following environment.
Our test program runs MQ Series 5.2 and BEA Weblogic 6.1sp2 on Solaris 2.8
over a E3500(4 cpus@400MHz, 3Gbytes of core memory). The application uses
the JMS API.
Tests showed that:
1.it takes about 150msec for a single thread to read a message from a MQ
Series queue in transactional context (using XA interface)
2.when using 5 threads to read the same queue, each thread takes about
400msec to read a message instead of keeping about the same 150msec
3.for a non-transactional read, a single thread is capable of reading a
message in about 50msec and the 5 thread version, in 70msec (still not
scalling up but not as bad as the transactional version).
Monitoring MQ using strmqtrc (MQ trace) at API level, we could see that
between xa_start and MQGET call it took about 300msec (using the 5 thread
version), apparently doing nothing.
CPU and I/O seems not to be a problem, at least as far as we could monitor.
Please see attached logs. The first log is the MQ trace file. The other one
is the app log which shows, for each thread, the time difference between one
message and the previous one to the same thread.
For instance
0milli (Thread-0)->Starting transaction <-before begin
transaction
1milli (Thread-0)->Waiting for message <-after begin tran,
before read queue
553milli (Thread-0)->Got incoming msg <-after read
1milli (Thread-0)->Transaction committing <-before commit
transaction
10milli (Thread-0)->Message processing completed. <-after commit
Are these figures the expected values for this hardware/software
configuration?
Regards.
[weblogic_ispb_20020123_163106.txt]
[trace.zip]

Answer 1

Make sure you are using the lwp thread library on Solaris 8 (put
/usr/lib/lwp FIRST in LD_LIBRARY_PATH - something like that).
I didn't look too closely at your complete post.
Mike
"Roberto Castro" <[email protected]> wrote in message
news:[email protected]...
Hi! I now MqSeries is not the subject of this newsgroup, but, as soon as,
I'm using Weblogic to access MQ JMS queue, I'd like to now the opinion of
other people about performance in the following environment.
Our test program runs MQ Series 5.2 and BEA Weblogic 6.1sp2 on Solaris 2.8
over a E3500(4 cpus@400MHz, 3Gbytes of core memory). The application uses
the JMS API.
Tests showed that:
1.it takes about 150msec for a single thread to read a message from a MQ
Series queue in transactional context (using XA interface)
2.when using 5 threads to read the same queue, each thread takes about
400msec to read a message instead of keeping about the same 150msec
3.for a non-transactional read, a single thread is capable of reading a
message in about 50msec and the 5 thread version, in 70msec (still not
scalling up but not as bad as the transactional version).
Monitoring MQ using strmqtrc (MQ trace) at API level, we could see that
between xa_start and MQGET call it took about 300msec (using the 5 thread
version), apparently doing nothing.
CPU and I/O seems not to be a problem, at least as far as we couldmonitor.
>
Please see attached logs. The first log is the MQ trace file. The otherone
is the app log which shows, for each thread, the time difference betweenone
message and the previous one to the same thread.
For instance
0milli (Thread-0)->Starting transaction <-before begin
transaction
1milli (Thread-0)->Waiting for message <-after begin tran,
before read queue
553milli (Thread-0)->Got incoming msg <-after read
1milli (Thread-0)->Transaction committing <-before commit
transaction
10milli (Thread-0)->Message processing completed. <-after commit
Are these figures the expected values for this hardware/software
configuration?
Regards.