Which machine should I use?


Your choice of machine should depend on the requirements and structure of your program.

Remember that single threaded software will only ever use a single CPU, so you should choose a machine with a fast processor. For software that can be mulithreaded, a higher number of CPUs will be beneficial. Multiple cores does mean that multiple users can run single threaded programs simultaneously.

Remember however that an older computer with fewer users may ultimately be faster than those with the best specs.

See the RTDC computer list


MIR/IDL
MIR/IDL is single threaded, this means it can only make use of one CPU. The biggest issue to consider is the memory requirement. MIR needs a considerable amount of RAM - typically 2-3 times the size of the dataset. This means our even machines with the biggest memories (rtdc8 and rtdc9) can only load in files up to ~65GB. You can work around this by reducing the size of your data file by rechunking using SMARechunker, or choose to only load a single receiver/sideband into MIR.

MIRIAD
For most processes, MIRIAD only uses a single CPU, but it does not have the same RAM requirements as IDL. MIRIAD benefits from a fast processor, and a large disk space if performing calibration steps. Disks with fast data exchange rates (ideally SSDs) are recommended, however we do not currently have any SSD disks on the RTDC network. At a minimum ensure you are not writing to a mounted disk.

CASA
CASA is partly multi-threaded so can make use of more than one CPU. However NRAO state that currently clock speed is more important than a large number of cores, and their 'highest-performance' recommendation only specifies 16 x 2.6GHz cores. As per the NRAO recommendations, rtdc9 has block level read-ahead set to 4MB.

The RAM requirements are moderate and, unlike IDL, not essential. If your task exceeds the available RAM it will slow down the processing but not stop it. The most memory intensive task is imaging, where NRAO recommend '4GB of memory per utilized processor core for spectral line and narrow field narrow band continuum imaging. For more memory intensive algorithms 8GB or 16GB per core may be required'.

Newer versions of CASA can be run parallelized by using MPI mode. This is not executed by default when running casa but requires running mpicasa and assigning a number of cores (or engines). In this mode data exchange is more important than ever so the fastest disks are recommended.

Read the NRAO recommendations here.

AIPS
AIPS is available on rtdc7, rtdc8, rglinux12, and rglinux13.