You can find here some useful information regarding Dhico usage, both in a generic sense and within the context of DiscoGrid.
Research papers:
Raúl López Lozano and Christian Pérez. – Improving MPI Support for Applications on Hierarchically Distributed Resources. In Recent Advances in Parallel Virtual Machine and Message Passing Interface, 14th European PVM/MPI User's Group Meeting, pp 187-194. – Springer Berlin / Heidelberg, Paris, France, 30 September - 3 October 2007.
-
User guide. This link provides a quick guide for using the runtime within an application:
from code changes to execution.
-
A tutorial explaining how to implement a simple data distribution module in DHICO.
A C API for taking advantage of the runtime.
There are two real applications in the DHiCo sources. Here are two manuals to run them.
A simple tutorial devoted to the explanation of two partitioner's
1) use cases and which files you need to run both
BHE and
CEM applications.
-
This section is aimed at gathering documentation used within the DHICO Runtime. This mainly includes:
-
-
Model considerations. This section deals with the model design and presents some core design decisions. (Please note that changes are being introduced in the runtime core, so not all aspects presented in this section are necessarily implemented on the same basis for the current version). The model includes some concrete features like:
This section contains some particular features of the model:
The Oracle. Find here some considerations about the oracle implementation.
-
-
SVN version 331 is '0.1' (APP)
SVN Version 117 is “stable”: It supports both MPI and sockets pcomm separately.
SVN Version 187 is “stable”: It supports both MPI and sockets at the same time (within the model constraints). This is the last version, with the old and slow socket configuration.
SVN Version 212 is “stable” version: This version has an improved socket configuration and connection under request implementation.
SVN Version 217 is the last “stable” version: Source files are slightly differently organized.
However it causes deadlocks when using the extra-big mesh (cf. falcon-2), as in this version all sends within the sockets are considered non-blocking, while they actually are if the send size is too big.
Dec. 2008: the problem comes from the FORTRAN application that is using large blocking sends. It needs to be modified!