Tom's explanation pretty much covers the high points. I don't do it yet 1) because it is not uniformly available, 2) in some scenarios folks really want a single master file writer model (putting the master where you have a good fast local disk), and 3) for any reasonable trajectory dumping frequency, this is not what currently kills us, as long as the performance characteristics of the file system are okay (and if they are not, then you don't solve the problem moving to mpi-io anyway). Between netCDF (primarily introduced by John Mongan) and some other stuff I have done to control write frequencies, we are actually in pretty good shape without too much complexity at the moment. Just because there is some technical "advance", there is not always a benefit to jumping on the bandwagon.
Regards - Bob Duke
----- Original Message -----
From: David LeBard
To: amber
Sent: Monday, June 11, 2007 4:58 PM
Subject: AMBER: Question about MPI-IO
Hi Everybody,
I am curious why AMBER does not use MPI-IO routines for writing trajectory/restart data? The only reason I can imagine would be in regards to the padding issues with unformatted fortran files and portability. Is this it? A detailed explanation would be great to settle my curiosity.
All the best,
David
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Wed Jun 13 2007 - 06:07:25 PDT
