Charm++ parallel programming system

["Kale, Laxmikant V" <kale AT cs.uiuc.edu>]

Yes, if the main chare is busy reading, and other depend on the data being read, they are going to be idle.

 

The current Charm++ viewpoint is that the programmer can use parallel file system (or parallel reads/writes) provided by the OS explicitly. NAMD, our MD program, is now being extended to do such parallel input. The unstructured mesh framework, for example, uses parallel input to read the partitioned mesh.

 

I don’t know what “support for parallel I/O” Abhinav had in mind; but one can imagine providing support for output/input of distributed data structures, and provide asynchronous I/O support; further ideas include an I/O interface that works in face of object migration. Some of these are in the plans for future research/effort.

 

--

Laxmikant Kale                  http://charm.cs.uiuc.edu

Professor, Computer Science     kale AT illinois.edu

201 N. Goodwin Avenue           Ph:  (217) 244-0094

Urbana, IL  61801-2302          FAX: (217) 265-6582

 

From: charm-bounces AT cs.uiuc.edu [mailto:charm-bounces AT cs.uiuc.edu] On Behalf Of Haowei Huang
Sent: Thursday, July 01, 2010 1:34 AM
To: Phil Miller; charm AT cs.illinois.edu
Subject: [charm] IO efficiency in Charm++

 

Hi Phil:

 

         I have a question about IO efficiency in Charm++. As shown in your manual, each program has a main chare and all the other chares are created by main chare through entry method. Take MD as an example, main chare reads all data information from input file and creates all patches and distributes data to different patches. In this situation, processor 0 is busy with reading data from file and other processors keep idle. Do you think is it a problem or do you have other asynchronous way to handle it? Thanks a lot.

--
Haowei Huang
Ph.D. student
Technische Universitaet Muenchen
Institut fuer Informatik, I10
Boltzmannstr. 3
D-85748 Garching
Room 01.06.061
Phone: +49 (89) 289 18477
mailto: huangh AT in.tum.de