Re: [ROOT] hsimple speed root_v3.04.02 vs root_v3.05.0x on Linux RH7.1-3

From: Rene Brun (Rene.Brun@cern.ch)
Date: Fri May 09 2003 - 09:20:12 MEST


Hi Andrzej,

I cannot reproduce this problem with $ROOTSYS/test/hsimple.cxx.

The main difference between 3.04 and 3.05 is the introduction by default
of the True Type Fonts. If you run the benchmarks.C with 3.04 without
using TTF and 3.05 with TTF, you may see a factor 2 on graphics
applications. If you use the new 3.05/05, you can disable TTF in system.rootrc.
We will work on the speed improvements with TTF.

Rene Brun

Andrzej Kupsc wrote:
> 
> Hej,
> 
>   We have found that some of our root based applications seems to run
> 2-4 times slower with root_v3.05.04 as compared to root_v3.04.02.
> 
> In order to find the reason of decreased performance we have run standard
> root benchmarks with precompiled root versions as well as root_v3.05.04
> compiled from sources.
> The comparison was done on the same computers with the only difference
> being root version used. Most dramatic difference is with
> $ROOTSYS/test/hsimple.cxx which seems to be more than 4 times slower with
> ROOT 3.05/04.
> Does anyone observed similar decrease of performance with root_v3.05.04
> or have hints what could be wrong?
> 
> Andrzej Kupsc
> ---------------------------------------------------------------
>                   The Svedberg Laboratory,
>                   Uppsala University, Uppsala.
> ---------------------------------------------------------------
> 
> --------------------
> Results of the tests
> --------------------
>  ROOT versions used for tests:
>   root_v3.04.02.Linux.RH7.2.gcc296.tar.gz
>   root_v3.05.04.Linux.RH7.3.gcc296.tar.gz
>     root_v3.05.04 recompiled from sources was also tested but that make
>     no differences with root_v3.05.04.Linux.RH7.3.gcc296
>     also we have tried root_v3.05.03 which behaves similar
> 
> 
> **1** Script benchmark.C is 2x slower for hsimple and
> hsum  we run it on our 2 computers (1 and 2 processor)
> with RH 7.1 (but results are similar on a computer with RH 7.3)
> 
>   computer1:  2x Pentium III
>   cpu MHz         : 799.693
>   kernel : 2.4.18-27.7.xsmp
>   Processing benchmark: rootmarks.C
> ---------------ROOT 3.04/02 benchmarks summary--------------------
> hsimple   : Real Time =   0.71 seconds Cpu Time =   0.65 seconds
> hsum      : Real Time =   0.31 seconds Cpu Time =   0.30 seconds
> .....
> ---------------ROOT 3.05/04 benchmarks summary--------------------
> hsimple   : Real Time =   1.32 seconds Cpu Time =   1.21 seconds
> hsum      : Real Time =   0.83 seconds Cpu Time =   0.83 seconds
> ....
> 
>   computer2:  1x Intel(R) Pentium(R) 4 CPU 1.80GHz
>   gcc version 2.96 20000731 (Red Hat Linux 7.1 2.96-98)
>   Linux: 2.4.7-10
> 
> Processing benchmark: rootmarks.C
> ---------------ROOT 3.05/04 benchmarks summary--------------------
> hsimple   : Real Time =   1.62 seconds Cpu Time =   0.89 seconds
> hsum      : Real Time =   1.10 seconds Cpu Time =   0.60 seconds
> ....
> Processing benchmark: rootmarks.C
> ---------------ROOT 3.04/02 benchmarks summary--------------------
> hsimple   : Real Time =   0.84 seconds Cpu Time =   0.47 seconds
> hsum      : Real Time =   0.42 seconds Cpu Time =   0.22 seconds
> ....
> 
> **2**
> Standard  $ROOTSYS/test/hsimple.cxx with standard Makefile was used.
> In order to test speed the event loop in the hsimple.cxx were increased to
> 1M events and TStopwatch was put at the beginning and at the end of the
> program to measure performance.
> 
> computer1:
>   ROOT 3.05/04
> ./hsimple
> Time : RT= 11.53 s  Cpu= 11.44 s
>   ROOT 3.04/02
> ./hsimple
> Time : RT=  2.20 s  Cpu=  2.19 s
> 
> computer2:
>   ROOT 3.05/04
> ./hsimple
> Time : RT=  7.88 s  Cpu=  7.67 s
>   ROOT 3.04/02
> ./hsimple
> Time : RT=  1.87 s  Cpu=  1.84 s



This archive was generated by hypermail 2b29 : Thu Jan 01 2004 - 17:50:11 MET