Обсуждение: Compile test with gcc 3.2

If you want to get the max (CPU) performance and use gcc, you should give the -fprofile-arcs / -fbranch-probabilties
optionsof gcc 3.2 a try.  
For 50 pgbench read-only runs (1 mio tuples, 40000 txs, 10 clients) I get 14.4% speedup.

Then I tried it with real data from our production system.
This is 2GB data, 120 tables, but most of the data is large object data (1.8GB), so most tables  of the database are
in-memoryand the application is more cpu bound. 
With this scenario, I still get 8% improvement.

All tests done on an Athlon XP/1500, 768MB RAM, Linux 2.4.19, gcc 3.2, 5400 RPM Maxtor.

Might be worth a try. Probably the performance win will be smaller for larger databases.

Regards,
    Mario Weilguni

Hi Mario:
--- Mario Weilguni <mweilguni@sime.com> wrote:
> If you want to get the max (CPU) performance and use
> gcc, you should give the -fprofile-arcs /
> -fbranch-probabilties options of gcc 3.2 a try.
> For 50 pgbench read-only runs (1 mio tuples, 40000
> txs, 10 clients) I get 14.4% speedup.
>
> Then I tried it with real data from our production
> system.
> This is 2GB data, 120 tables, but most of the data
> is large object data (1.8GB), so most tables  of the
> database are in-memory and the application is more
> cpu bound.
> With this scenario, I still get 8% improvement.
>
> All tests done on an Athlon XP/1500, 768MB RAM,
> Linux 2.4.19, gcc 3.2, 5400 RPM Maxtor.
>
> Might be worth a try. Probably the performance win
> will be smaller for larger databases.

 - Do you use the "-fprofile-arcs
-fbranch-probabilties" options with other optimization
flags? I've in a book (Optimimizing Red Hat Linux 6.2)
that one could also optimize speed by setting the
CFLAGS to the following :
  " -02 -fomit-frame-pointers -funroll-loops" and
  running "strip" on the on binaries after they are
compiled.

regards,
ludwig.

__________________________________________________
Do you Yahoo!?
New DSL Internet Access from SBC & Yahoo!
http://sbc.yahoo.com

except when you run the regression tests you get some very interesting results.

On Fri, 11 Oct 2002 07:34:58 -0700 (PDT)
Ludwig Lim <lud_nowhere_man@yahoo.com> wrote:

>
>  - Do you use the "-fprofile-arcs
> -fbranch-probabilties" options with other optimization
> flags? I've in a book (Optimimizing Red Hat Linux 6.2)
> that one could also optimize speed by setting the
> CFLAGS to the following :
>   " -02 -fomit-frame-pointers -funroll-loops" and
>   running "strip" on the on binaries after they are
> compiled.
>