[Carpet] PsiKadelia and Carpet

[Erik Schnetter]

On Nov 13, 2007, at 18:04:57, Hee Il Kim wrote:

> Hi,
>
> I have a very unclear question. When I use PsiKadelia with Carpet,  
> I encounter some MPI related errors after about 100 timesteps. At  
> first, I thought this might be nfs related network errors but this  
> happens even I remove any IO parameters. This error happens if I  
> set the followings with Carpet,
>
> ### psikadelia
> psikadelia::ricci_persists = yes
> psikadelia::weyl_persists = yes
> psikadelia::ricci_timelevels = 3
> psikadelia::weyl_timelevels = 3
>
> I don't see any errors yet, if I use PsiKadelia with PUGH. Does  
> PsiKadelia contain any high level of MPI functions which might  
> cause errors on a humble cluster?. I use 5 node diskless ramdisk  
> cluster with nfs server.

Hee Il,

PsiKadelia contains approximately the same MPI complexity as one  
right hand side evaluation of BSSN.  It's not particularly more complex.

What kind of error messages do you get?  Are these errors where MPI  
complains that there is an error in the programme, e.g. a deadlock,  
or is it an error where something within MPI has become inconsistent,  
e.g. a lost message?  Does the problem also appear when you run on a  
single node?

If you suspect a hardware problem, you can go to <http:// 
www.hpc.lsu.edu/accounts/> and apply for an account at the LSU HPC  
machines, naming me as sponsor.  You can then run the same code and  
parameter file on one of the LSU machines and see whether the error  
persists.

-erik

-- 
Erik Schnetter <schnetter at cct.lsu.edu>

My email is as private as my paper mail.  I therefore support encrypting
and signing email messages.  Get my PGP key from www.keyserver.net.



-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 186 bytes
Desc: This is a digitally signed message part
Url : /archives/developers/attachments/20071113/77b0207d/attachment.pgp