[Paraview] Running ParaView with MPI

Tue, 11 Feb 2003 16:25:54 -0600

This is a multi-part message in MIME format.

------=_NextPart_000_00C2_01C2D1EA.403B0F40
Content-Type: text/plain;
	charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable

Thanks for the reply Berk. My issue turned out to not be a problem after
all. Apparently, newer Xeons and Linux kernels support hyperthreading, =
so
each node in my cluster is actually reporting (/proc/cpuinfo) 4 logical
processors (2 physical) to MPICH and ParaView, thus the 25% maximum cpu
usage/process.

I am curious if anyone has benchmarked ParaView with/without =
hyperthreading.
Does anyone have any recommendations as to whether it should be on or =
off?

----- Original Message -----
From: "Berk Geveci" <berk.geveci@kitware.com>
To: "Steven Coorpender" <scoorp@lsuhsc.edu>
Cc: <paraview@public.kitware.com>
Sent: Friday, February 07, 2003 2:33 PM
Subject: Re: [Paraview] Running ParaView with MPI

> ParaView doesn't make any calls to require system resources nor
> does it ever ask to be run a certain processor (it can't anyway,
> there is no such standard MPI call). Therefore, it will use whatever
> the operating system and/or the MPI library gives it. Of course,
> unless you are doing some expansive computation, there is no reason
> why ParaView should use all the processor time available. It will
> idle until it is asked to compute something. This might be true
> even if you are interacting with an object since, when using hardware
> accelerated rendering, the graphics processor will take the
> actual load.
>
>
> On Fri, 2003-02-07 at 15:13, Steven Coorpender wrote:
> > I am running ParaView with MPICH v1.2.5 on a 2-node cluster (each =
node
is a dual processor Xeon 2.2GHz running RedHat 8.0-smp--total of 4
processors in cluster). When I run ParaView with "mpirun -np 4 =
ParaView", my
system monitor shows a maximum of 50% cpu usage on each node, giving 25% =
to
each ParaView process. If I run Paraview with "mpirun -np 8 ParaView", I
still get 25% for each process, but 4 processes on each node, and a =
maximum
of 100% total cpu usage. Why won't ParaView use more than 25% of a cpu =
in
one process, and why can I only get 50% usage with mpirun -np 4?
>
>
>

------=_NextPart_000_00C2_01C2D1EA.403B0F40
Content-Type: text/html;
	charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML><HEAD>
<META http-equiv=3DContent-Type content=3D"text/html; =
charset=3Diso-8859-1">
<META content=3D"MSHTML 6.00.2800.1126" name=3DGENERATOR>
<STYLE></STYLE>
</HEAD>
<BODY bgColor=3D#ffffff>
<DIV>Thanks for the=20
reply Berk. My issue turned out to not be a problem after all. =
Apparently,=20
newer Xeons and Linux kernels support hyperthreading, so each node in =
my=20
cluster is actually reporting (/proc/cpuinfo) 4 logical processors (2 =

physical) to MPICH and ParaView, thus the 25% maximum=20
cpu<BR>usage/process.<BR><BR>I am curious if anyone has benchmarked =
ParaView=20
with/without hyperthreading.<BR>Does anyone have any recommendations as =
to=20
whether it should be on or off?<BR><BR>----- Original Message =
-----<BR>From:=20
"Berk Geveci" &lt;</FONT><A =
href=3D"mailto:berk.geveci@kitware.com"><FONT=20
face=3D"Times New Roman" =
size=3D3>berk.geveci@kitware.com</FONT></A><FONT=20
face=3D"Times New Roman" size=3D3>&gt;<BR>To: "Steven Coorpender" =
&lt;</FONT><A=20
href=3D"mailto:scoorp@lsuhsc.edu"><FONT face=3D"Times New Roman"=20
size=3D3>scoorp@lsuhsc.edu</FONT></A><FONT face=3D"Times New Roman"=20
size=3D3>&gt;<BR>Cc: &lt;</FONT><A =
href=3D"mailto:paraview@public.kitware.com"><FONT=20
face=3D"Times New Roman" =
size=3D3>paraview@public.kitware.com</FONT></A><FONT=20
face=3D"Times New Roman" size=3D3>&gt;<BR>Sent: Friday, February 07, =
2003 2:33=20
PM<BR>Subject: Re: [Paraview] Running ParaView with MPI<BR><BR><BR>&gt; =
ParaView=20
doesn't make any calls to require system resources nor<BR>&gt; does it =
ever ask=20
to be run a certain processor (it can't anyway,<BR>&gt; there is no such =

standard MPI call). Therefore, it will use whatever &gt; the =
operating system=20
and/or the MPI library gives it. Of course, &gt; unless you are doing =
some=20
expansive computation, there is no reason &gt; why ParaView should =
use all=20
the processor time available. It will &gt; idle until it is asked to =
compute=20
something. This might be true &gt; even if you are interacting with =
an object=20
since, when using hardware &gt; accelerated rendering, the graphics =
processor=20
will take the &gt; actual load. &gt; &gt; &gt; On Fri, =
2003-02-07 at=20
15:13, Steven Coorpender wrote: &gt; &gt; I am running ParaView with =
MPICH=20
v1.2.5 on a 2-node cluster (each node is a dual processor Xeon 2.2GHz =
running=20
RedHat 8.0-smp--total of 4 processors in cluster). When I run =
ParaView with=20
"mpirun -np 4 ParaView", my system monitor shows a maximum of 50% cpu =
usage=20
on each node, giving 25% to each ParaView process. If I run Paraview =
with=20
"mpirun -np 8 ParaView", I still get 25% for each process, but 4 =
processes on=20
each node, and a maximum of 100% total cpu usage. Why won't ParaView =
use more=20
than 25% of a cpu in one process, and why can I only get 50% usage =
with=20
mpirun -np =
4? &gt; &gt; &gt; </DIV></BODY></HTML>

------=_NextPart_000_00C2_01C2D1EA.403B0F40--