too much cpu usage

[inactive account]

Our applications on HPUX 11.00 uses 7 times more CPU than on an intel PIII 500 with LINUX.
Everything is written in C++ and we use aCC on HP as the compiler.
It seems that the STL are responsible of the over-usage.

Does anybody know a solution or a way to decrease CPU usage .

PS: I use optimize level 2 only because other levels do not work while I'm using a lot of throw /catch

Thanx

[Steve Steel]

Hi


1)Check your patch level for cpu issues

Example

Patch Name: PHSS_13391

Patch Description: s700_800 11.00 HP aC++ (A.03.05) to fix numerous defects

Creation Date: 97/12/17

Post Date: 98/01/16

Hardware Platforms - OS Releases:
s700: 11.00
s800: 11.00

Products:
HP aC++ B3910B A.03.04

Filesets:
ACXX.ACXX,fr=B.11.00,fa=HP-UX_B.11.00_32/64,v=HP
ACXX.ACXX-MAN,fr=B.11.00,fa=HP-UX_B.11.00_32/64,v=HP

Automatic Reboot?: No

Status: General Release

Critical: No

Category Tags:
defect_repair general_release

Path Name: /hp-ux_patches/s700_800/11.X/PHSS_13391

Symptoms:
PHSS_13391:
1. Excessive CPU and VM usage and compile time with STL


2)PRM will do it


3)HP machines will use CPU if it is available.
Do you see a problem with the excessive cpu

Regards

Steve Steel

Quote of the moment
-------------------
"We are drowning in information but starved for knowledge."
-- John Naisbitt

[Dietmar Konermann]

Hi!

It's really a difficult thing to compare CPU usage on different systems!

Imagine a programm that does some internal calculation, performs IO, calculates again, performs IO and so on.

Imagine machine 1, where the calculation and the IO task take 5 seconds each... on this machine you will see 50% CPU usage.

Now imagine machine 2, same CPU power, but the IOs are twice as fast... here you will see about 66% CPU usage (5 secs for each calculation and 2.5 secs for each IO).

And what is the conclusion? Would it be bad to see more CPU usage here? Of course not.

Best regards...
Dietmar.

PS:

Always remember Performance Tuning Rule #1:

"When answering a question about computer sytem performance, the initial answer is always: It depends."

:-)

[inactive account]

Yes it is definitely an issue because when you can run 50 transactions per second on a simple PC you should normally expect to have at least the same performance on a PA RISC 8500 !

And when the cpu time if 100% used you can't expect to increase the ratio . THe fact is that now I poorly can run 10 transactions per second where I should have 25...

COncerning the patch level of the compiler : we are working with version 03.35 of aCC

[Steve Steel]

Hi


Go there
http://h21007.www2.hp.com/dspp/tech/tech_TechSoftwareDetailPage_IDX/1,1703,1740,00.html

You will see that there is a later compiler and rogue wave stl


try that

Steve Steel

[Martin Johnson]

Have you attempted to tune your HPUX system?

Are your application processes getting enough memory? Are your I/Os being properly cached?
Are your I/Os spread over more than one spindle? Have you checked what other processes are running on the system and made sure they are not in contention with the applications? Have you insured that the applications are not in contention with themselves?

HTH
Marty

[Dave Wherry]

I doubt that any one can give you an answer with just the information that it uses more CPU. Performance is always about more than a single metric.
There are situations where a single process using 100% CPU is fine. If it is not contending with any other processes. Let it have it all. If it's starving other processes then there is a problem.
You have to consider memory, I/O, networking...

[Steve Steel]

Hi

As I said before get the later compiler aCC A.03.37 and add this patch

Patch Name: PHSS_26945

Patch Description: s700_800 11.X HP aC++ -AA runtime libraries (aCC A.03.37)

Creation Date: 02/09/25

Post Date: 02/10/28

It solves a lot of roguewave stl problems


Regards

Steve Steel

Quote of the moment
-------------------
"We are drowning in information but starved for knowledge."
-- John Naisbitt

[A. Clay Stephenson]

You are really not providing enough data to begin to solve this. The first thing that I would do is compile using -G and then examine the executable using gprof. Profiling will enable you to better understand what your code is spending much of its time doing. Another thought that I have is to check your setting of timeslice; if someone has set it to 1 rather than 10 you might very well see exactly this situation.

[inactive account]

We tried using STLport instead of roguewave the gain is evident!