5.5 Migration Toward Grid Environments

It is useful, at this point of our analysis, to make a short summary of what we have discussed up to now. As seen in the previous sections, the CAE of rectangular-aperture arrays can be fruitfully partitioned into four main subproblems, namely:

1. AFS;

2. AMC;

3. ESM;

4. ERP.

Each subproblem can correspond to a single, independent module. Each module can in principle be executed independently from the others, provided that suitable input data are available.

Each item needs specific numerical approaches and has peculiar properties and difficulties. Each item requires more or less relative computational effort. Points 2 and 3 are particularly CPU-intensive applications and can take advantage of the use of HPC platforms. It has been demonstrated that the use of a multilevel parallelism, implemented in SPMD programs with MPICH, is a viable pathway to achieve very effective performance on MIMD platforms. In order to maintain the maximum flexibility of the package, a modular structure is still adopted in the parallel implementation, with four independent building blocks.

In accordance with what we've now recalled, the problem of migrating a CAE framework for aperture arrays toward GC is a little more complicated with respect to the case of parallel FDTD (see Chapter 4). In fact, we still need to support an HPC demand, but must fulfill this requirement with an additional and relevant demand for interoperability among several (namely, four) independent, heterogeneous, and geographically distributed modules.

In Chapter 4, we described how a generic parallel application, developed...