Performance problems on RedHat Linux 2.4.21 kernel

gasproni

Hi all,

I'm in the process of evaluating ICE to propose it as a middleware solution to be adopted in my current project.

I run some tests under windows 2000 and everything was fine. In the last couple of days I wanted to run some tests using some high performance RehHat linux machines (4 2.GHz processors with hyperthreading), and gcc 3.2.3; and I had some problems.

The first one was that the compilatrion halted because it couldn't find a definition for the DB_LOG_AUTOREMOVE macro while compiling Freeze. Since I didn't need Freeze, I put a definition to 1. The compilation then ended successfully (all the tests except the Freeze one passed).

However, when I run the tests the performance were incredibly low--an order of magnitude less than much smaller windows machines.

I was wondering if I missed something obvious, if not I can even post the c++ code I used for the tests.

Thanks,
Giovanni

gasproni

The processors are 2.8 GHz

Giovanni

gasproni wrote:

Hi all,

I'm in the process of evaluating ICE to propose it as a middleware solution to be adopted in my current project.

I run some tests under windows 2000 and everything was fine. In the last couple of days I wanted to run some tests using some high performance RehHat linux machines (4 2.GHz processors with hyperthreading), and gcc 3.2.3; and I had some problems.

The first one was that the compilatrion halted because it couldn't find a definition for the DB_LOG_AUTOREMOVE macro while compiling Freeze. Since I didn't need Freeze, I put a definition to 1. The compilation then ended successfully (all the tests except the Freeze one passed).

However, when I run the tests the performance were incredibly low--an order of magnitude less than much smaller windows machines.

I was wondering if I missed something obvious, if not I can even post the c++ code I used for the tests.

Thanks,
Giovanni

marc

Which test or demo gave you bad performance? Did you compile with optimization? Which compiler flags did you exactly use?

marc

Also, which version of Ice are you using?

gasproni

marc wrote:

Also, which version of Ice are you using?

Hi Marc,

The test is one we've written internally, the ICE version is 2.1.2, and we used -DNDEBUG, and no -Ox optimization flags.

Anyway, yesterday we spotted the problem: in that specific platform (Linux with gcc 3.2.3) we wrote a call

IceUtil::ThreadControl::sleep( IceUtil::Time::milliSeconds( sleepMillis_ ) );

This function was behaving strangely when sleepMillis_==0: it actually made the thread sleep for one second. We commented it out and everything was fine.

I think it may be a bug in the compiler.

Giovanni

marc

If you measure performance, it's very important that you use -O2 or -O3. Without optimization enabled, Ice is much slower, in particular because of the STL.

gasproni

ICE was compiled out of the box with optimization turned on. Our code wasn't, however the performances were amazing :-)

Giovanni

bernard

Regarding the DB_LOG_AUTOREMOVE problem, it's probably because the version of Berkeley DB you have installed is too old.
DB_LOG_AUTOREMOVE was introduced in Berkeley DB 4.2; Ice needs 4.2 or 4.3.

Bernard