XANSYS Message: 3268 [Go back to message list]
[bookmark on del.icio.us]
No rating yet
Rate item:

Subject: Re: MultiThreading on a PC
Author: Patrick J. Sheehan pat@
Date: 1999-07-14 10:05:00

Chris and all,

Block Lancos is not parallelized on SGI machines under ANSYS 5.4. In fact,
one must include a line their .cshrc file to defeat multi-threading on a
multi processor
machine in order to run Block Lancos. I even had problems running static
models
with the sparse solver until multi-threading was defeated.

We were informed that it may be in Beta release at ANSYS 5.6.

Chris Mollo wrote:

> From: "Chris Mollo"

> I had a user call-in about a very similiar problem...
> ANSYS 5.5.x would "hang" when he used more than 1
> processor on an Intel-based PC with NT 4.0. It did
> not abort; it would just grind to a halt, i.e. CPU
> usage would tend towards zero. "Small" models ran
> fine; "large" ones had the problem. I can't remember
> what type of analysis he was doing (and I'm too lazy
> to look it up).

> Anyhow, the engineers at ANSYS, Inc. debugged the
> problem and the fix is in ANSYS 5.6.

> According to the engineers at ANSYS, Inc. this problem
> only occurs when you run ANSYS in interactive mode.
> There is no problem running in batch mode. Have you
> tried running in batch mode? If not, give it a try.
> Otherwise, your only other workaround is to set NPROC
> to 1. (There is no parallel processing for NPROC=1,
> to answer another XANSYS posting.)

> One caveat to John Swanson's reply to the question if
> the Block Lanczos solver is parallelized: it is, at
> least to a certain extent, on 64 bit SGI machines
> since ANSYS 5.4. The sparse solver is parallelized on
> 64 bit SGI machines and the Block Lanczos solver uses
> the sparse solver. On all other platforms, as John
> Swanson said, the Block Lanczos solver does not yet
> support parallel processing.

> Chris Mollo
> Automated Analysis Corp.
> chris@a...

> From: Weathers, Bob
> To: 'xansys'
> Subject: MultiThreading on a PC
> Date: Monday, July 12, 1999 7:56AM

> 1. Running NT 4.0 (SP4 or SP5)
> 2. Using ANSYS 5.5.2 or 5.5.3
> 3. Solving a large (~600K DOFs) modal using PowerDynamics or Block
> Lanczos solvers
> 4. Using '/CONFIG, NPROC, 2' always caused the solution to abort before
> fully solved. (Tried it 6 times on 2 PCs)
> 5. Omitting the preceding /CONFIG command lets a solution be obtained.

> Anyone else seen this behavior on a PC?

> Bob Weathers
> bweathers@t...
> (608)787-2729

Posts possibly associated with message #3268AuthorDateScore
3217MultiThreading on a PCBob Weathers1999/07/12 
3222Re: MultiThreading on a PCMike McNamee1999/07/12 
3223Re: MultiThreading on a PCBob Weathers1999/07/12 
3225Re: MultiThreading on a PCCosme, Carlos CCosme@1999/07/12 
3229Re: MultiThreading on a PCJohn Swanson1999/07/12 
3238Re: MultiThreading on a PCChris Mollo1999/07/13 
3239Re: MultiThreading on a PCBob Weathers1999/07/13 
3251Re: MultiThreading on a PCDoug Scott1999/07/13 
3265Re: MultiThreading on a PCChris Mollo1999/07/14 
3268Re: MultiThreading on a PCPatrick J. Sheehan pat@1999/07/14 
3270Re: MultiThreading on a PCBob Weathers1999/07/14