PostgreSQL7.1 on AIX5L is running with too poor ferformance
Hi
PostgreSQL7.1 is now running on AIX5L( S85, 6GB memory, 6CPU), which was
running on Linux before(Pentium3, 2CPU, as far as I
remember.......sorry......).
The performance( on AIX5L ) is just half as good as the one( on Linux ).
I compiled PostgreSQL on AIX5L ofcourse.
I haven't configured it when migrating to AIX5L though.
Are there any problems in not tuning when migrating it to AIX5L?
What should I check first?
I can't make a head or tail of it:(
Help!!
Best regards,
Shuichi
"Leslie" <leslie@boreas.dti.ne.jp> writes:
PostgreSQL7.1 is now running on AIX5L( S85, 6GB memory, 6CPU), which was
running on Linux before(Pentium3, 2CPU, as far as I
remember.......sorry......).
The performance( on AIX5L ) is just half as good as the one( on Linux ).
Hmm ... is the AIX compilation selecting an appropriate TAS
implementation for spinlocks? If it's falling back to semaphore-based
spinlocks, I can easily believe that you might take a 2X performance
hit. Possibly s_lock.h needs some additional #if tests for AIX5.
regards, tom lane
"Leslie" <leslie@boreas.dti.ne.jp> writes:
PostgreSQL7.1 is now running on AIX5L( S85, 6GB memory, 6CPU), which was
running on Linux before(Pentium3, 2CPU, as far as I
remember.......sorry......).
The performance( on AIX5L ) is just half as good as the one( on Linux ).Hmm ... is the AIX compilation selecting an appropriate TAS
implementation for spinlocks?
I think yes. I have compiled 7.1 on an AIX5L box and found that TAS()
was replaced by:
cs((int *) (lock), 0, 1)
If it's falling back to semaphore-based
spinlocks, I can easily believe that you might take a 2X performance
hit. Possibly s_lock.h needs some additional #if tests for AIX5.
--
Tatsuo Ishii