HP-UX Itanium 9320 vs RHEL performance

[ATveretinov]

Hi,

We bought a new HP-UX RX2800 server with Itanium 9320 with hope to increase our application performance and move from RHEL.

But simple OpenMP test shows downgrade of performance. Even with bigger number of threads. Simple parallel loop is very slow on HP-UX. Time = 6 secs. On RHEL this test takes about 1 sec.

As I understand the HP-UX should be faster using more cores.....but no :(

All optimization for HP-UX aCC compiler is on.
Code:

for(size_t j = 0; j < N; ++j)
{
                #pragma omp parallel for default(shared) private(i) schedule(static,chunk) reduction(+:result)
                for (ssize_t i = 0; i < N; ++i)
                {
                    result = result + (a[i] * b[i]);
                }
}

Compiller string:  aCC -AA -mt +Ofaster -ipo +Ofltacc=relaxed +Onosize +Oinline +DO11.31 +DSnative +DD64 +Oopenmp

Is anyone know the problems with this? Any suggestions?

Thanks for help.
Alex.

[Dennis Handly]

What hardware is RHEL running?

How big is N?  What are the types of a and b?

What version of aCC6 are you using?

[ATveretinov]

RHEL has 4 cores 2.2 Hz

No matter how big is N. In my example N=5000.

a and b are 2 arrays with integers.

aCC: HP C/aC++ B3910B A.06.26 [Apr 12 2011]

[ATveretinov]

I just tested if all threads work with shared data - time on HPUX is bigger, if we change the test and make the result variable private - time on RHEL and HPUX will be almost the same.

[Dennis Handly]

Perhaps using +Oinfo may tell us something?

[ATveretinov]

Ye, the compiler said how I can improve the aCC STD library :))

Strange. Pay like for Ferrary and ride using horse :((