Features of parallel PHOENICS running

If a user wants to make effective parallel calculations then he should be sure that he has good PC-cluster. With the advent of cheap multi-core PCs, this task has been somewhat simplified as, one can often obtain sufficient parallel computing power within a single box. If one needs to set up a traditional cluster using multiple PCs then they should be connected via a gigabit ethernet switch and be configured to enable TCP/IP socket connections between all computers. For efficient parallel computations, it is preferable to have all PCs in the cluster of the same specification.

It is difficult to make specific recommendations of hardware, as these would depend on the budget available and the preferences of your organisation. PHOENICS in general, and the solver in particular, does not require any special graphics capabilities, so in chosing a PC we would recommend spending the extra on faster processors and additional RAM. Users seem to want to run larger and larger models, so it is important to have sufficient RAM available to run these models. On a multi-core PC, with a 32 bit operating system, we would recommend having upto two gigabytes of RAM for each processor, plus and addition one gigabyte of RAM to handle the operating system. If the user has a 64 bit machine and operating system then the user may extend beyond this.

When a user connects his PCs on the local network, he hopes to get accelerations in his calculations using parallel program. Often this is the case, but sometimes he sees deceleration. To understand the reasons for deceleration the user should know something about parallel programming. Each parallel CFD program uses decomposition of domain into several sub-domains. The usual number of sub-domains is equal to the number of processors in PC-cluster. Each sub-domain exchanges data with other sub-domains. If many data items are exchanged, then processors will spend much time in exchanging them and useful time for calculations will be very small. This situation demonstrates ineffective use of parallel calculations. Effective parallel calculations will result when the time of calculations is significantly more than the time of data exchange. Hence users should use decompositions which have smaller amount of cells on boundaries. Time of data exchange is connected with latency and time for single exchange which are characteristics of the PC-cluster.

Running of parallel PHOENICS has the several features:

  • Very popular slab-wise solution of hydrodynamics tasks should be abandoned because it leads to a lot of exchanges of small data instead of one exchange of whole data which whole-field manner is provided. User should inspect the file q1ear to sure that all variables are run whole-field. It is especially concerned the models of turbulence which use slab-wise solution by default, for example, statement TURMODE(KE) in Q1 leads to slab-wise solution of several variables. Parallel run of such case is non-effective.
  • Usually automatic decomposition gives the best decomposition. But sometimes user wants to use non-standard decomposition or he want to run case with Multi-Block grid. In this case he should to write file PARDAT manually. Description of file PARDAT see here.
  • User should remember that some features of sequential PHOENICS are not implemented in parallel. For example, InForm in sequential PHOENICS allows to link several cells by formulae, parallel PHOENICS does not allow it.