Test+Problem+Running

=Running the Test Problems= Test problem pages: Overview, CADAC, Data Access, Utilities, Work Area Setup, Path Setup, Grid Data, SPH Data, **Running**, Reporting

You need to run the test problem on a system where you have access to sufficient computational resources. CADAC accounts at SDSC are for data access and analysis, not for production runs.

Transferring initial value files
If you plan to run at 1024^3 or more it is probably most efficient to transfer the original (1000^3) data file, and construct your initial conditions locally (e.g. using a copy of the Test Problem Utilities). If you are planning to run at 512^3 or less it is probably more efficient to construct the initial conditions at SDSC, and transfer the (smaller) initial data files.

Transferring results back to SDSC
As per the Test Problem Overview page we ask you to transfer your results back to SDSC, and store them on the srcVault (see the Data Access page). In order to facilitate the use of standard tools to evaluate and compare the data sets it is a good idea to translate the data back to a raw data format of the type produced by the Test Problem Utilities. Alternatively, if you transfer the files in a different format, you need to supply an I/O interface that allows them to be read at SDSC.

Transferring data from and to SDCS
To transfer data from and to SDSC you may find the [|TeraGrid tools and documentation] useful. These are utilities that allow a faster (in some cases **//much//** faster!) transfer rate, and with retry of transfer until it succeeds. In general these utilities work by opening up more than one transfer channel (thread) at a time. If, for some reason, you cannot use the TeraGrid tools (e.g. lack of suitable node or client), you can increase transfer speed by simply splitting your dataset into many pieces (this is perhaps in any case part of your MPI setup), and start a number of //scp// commands in parallel.