Newsgroups: comp.parallel.mpi
From: Alexandre Carissimi <Alexandre.Carissimi@imag.fr>
Subject: MPICH x LAM performance
Organization: LMC-Laboratoire de Modelisation et Calcul
Date: Fri, 18 Apr 1997 13:59:44 +0200
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <33576230.746A@imag.fr>

I'm doing some performance measures using MPI. Nowadays
I'm using MPI LAM (version 6.1) and MPICH (version 1.0.14). The
MPICH version is configured to use p4 as channel device. My test
platform is composed by:
        - DELL OPTIPLEX GXL 5133 
        - mother board pentium 133 Mhz (intel)
        - Bios on flash eprom
        - 256 kbytes cache memory (pipe line burst)
        - 64 Megabytes RAM
        - HD 2 Gigabytes EIDE (Seagate ST32140A)
        - Chipset to networking "on board" 3COM Ether 3C913
        - Board FAST ethernet 3C595
	- Myrinet network interface PCI
	- Operating System: SOLARIS 2.5.1

	First of all using "netperf" I found a latency time equals
to:
	- 327 us using the fast ethernet interface
	- 420 us using the myrinet (its seems for me, normal because
          myrinet has a kind of TCP/IP emulation)

	Now, using LAM 6.1, the latency is:
	- 350 us using the fast ethernet interface
	- 480 us using the myrinet

	(LAM is executed by mpirun -c2c -nger -w -O options)

	Using MPICH, and doing exactly the same procedures, the
results obtained are:
	- 700 us using the fast ehernet interface
	- 700 us using the myrinet interface.

	(MPICH was installed with -nondevdebug option)

	I'm sure that I'm using the correct interface because I monitored
the interfaces using netstat command. My questions are:

	. Its normal this "overhead" de MPICH if compared with LAM
and NETPERF values. Why the latency is almost twice time? Is the
P4 level so heavy??
	. OR MPICH/P4 has any kind of deamon which pools the network
or the OS masking the real latency time??

	Ideas, suggestions, comments??

	Thanks in advance.

		ASC
__________________________________________________________________
CARISSIMI, Alexandre                  Alexandre.Carissimi@imag.fr
Laboratoire de Modelisation et Calcul - LMC
Institut National Polytechnique de Grenoble - INPG
100, Rue des Mathematiques               Fone: (33) 04.76.51.46.03
38041 - Grenoble - Cedex 9  FRANCE       Fax : (33) 04.76.51.48.46
                            (International calls removes 0 from 04)

"Le seul moteur pour gagner c'est innover" Renault (F1 - 1996)
________________________________________________________________