Suggested future work

1. The performance of Grids like the MecaGRID is believed to be best when all the processors have public IP addresses avoiding the need for VPNs and tunneling. Thus another extremely important motivation for integrating INRIA Rhone-Alpes as an active member is to permit evaluate the tunneling VPN and the public IP address approaches to determine relative performance and to establish at which point the efficiency of tunneling starts to deteriorate.

2. Another reason for integrating the INRIA Rhone Alpes cluster(s) into the MecaGRID is to take advantage of the existing 2.5 Gbps VTHD high-speed network connecting INRIA Sophia Antipolis and INRIA Rhone Alpes that should result in an efficient Grid.

3. The shockwave-bubble test case is time evolutionary and one is obliged to save the solution profiles at certain intervals in order to create animations. Saving solutions increases the global communication times if MPI sends/receives are used to gather the subdomain profiles on processor "0". This should be avoided whenever possible. A suggestion to minimize global communication times would be to write the profiles locally on each processor and recover them after the run has been completed using scp or sftp.

4. Create global submittal scripts using the Globus ToolKit MetaComputing Directory Service.

5. Upgrade the Globus software at regular intervals 8-12 months minimum.

6. Hardware upgrades at the different sites should be made taking into account Grid computing.

7. Global communication times increase significantly for inter-cluster executions. AEDIF is a research code with many global communications that could be eliminated to improve inter-cluster performances. For example, one could eliminate computing the maximum and minimum of density, pressure, u, v, w.

8. The role of mesh decomposition needs to be investigated.