As described in the page on running DiFX you need to provide mpifxcorr with machines and threads files and an appropriate mpirun command. There are utilities to automate this. Currently the primary options are genmachines/startdifx, espresso and startcorr.pl.
The tools above all need a basic definition of your cluster in order to automatically set up the required MPI files. A standard cluster definition file format has been agreed to facilitate this, and will be used by all the above tools starting from the DiFX-2.3 release.
# is used to indicate a comment. The
# and all subsequent characters on the line are not parsed. Empty lines and comment-only lines are permitted (and ignored).
The first line of the file gives the version of the cluster file format in the form
version = <INTEGER>. This version number is to differentiate between different versions of the cluster definition file format, in case changes to the format are required in the future. Currently only
version = 1 is valid.
Subsequent lines in the file define the nodes that comprise the cluster. Each node definition line contains a comma separated list of values that define the relevant features of one or more nodes. If a node name appears more than once in the file, the last entry for that node is used (later entries supersede earlier ones). The meanings and allowed formats of each column are given in the following table:
|1||String, with zero-padded numeric expansion of numbers contained in square brackets. E.g. cuppa[01-20] expands to include nodes cuppa01, cuppa02, cuppa03 … cuppa20. For expansions within the square brackets, the number of digits must be the same (with zero padding used where appropriate). Only one set of square brackets can be used on a single line.||The name of the cluster node(s). Multiple identical nodes can be described by a single line using the '[ ]' notation.|
|2||integer||0 means the node is disabled, 1 means the node is enabled, 2 means the node is enabled and eligible to be the master node|
|3||integer||the maximum number of compute threads to be used on this node. Nodes which are only to be used as datastream or master nodes may be given the value 0 to avoid using them for compute|
|4|| space separated urls. ||a space separated list of urls for data sources on this node. Allowed url formats are: file://<path> for directories containing files, mark5:// for Mark5 machines, network://<IP> for an eVLBI data source (<IP> is the external IP address of the node), mark6: // for Mark6 machines|
An example cluster definition file:
version = 1 # version number is an integer # node, enabled/disabled (2=head node), number compute threads, space separated list of urls for data # Later lines supersede earlier lines cuppa01, 2, 7, file:///exports/xraid01/l_1/corr file:///exports/xraid01/r_1/corr # possible master node cuppa02, 1, 7, file:///exports/xraid02/l_1/corr file:///exports/xraid02/r_1/corr cuppa03, 1, 7, file:///exports/xraid03/l_1/corr file:///exports/xraid03/r_1/corr file:///arch/corr/bbdata/corr cuppa05, 1, 7, file:///exports/xraid05/l_1/corr file:///exports/xraid05/r_1/corr cuppa[07-15], 1, 7 # zero-padded numeric ranges are allowed cuppa12, 0, 7 # disable cuppa12 - supersedes previous line in expanded range. cuppa16, 1, 7, file:///mnt/disk1/corr file:///mnt/disk2/corr cuppa17, 1, 7, file:///mnt/disk1/corr cuppa18, 1, 7, file:///mnt/disk1/corr cuppa19, 1, 7 cuppa2[1-4], 1, 0, file:///mnt/raid/corr # datastream only nodes - 0 compute threads. mark5b-1, 1, 0, mark5:// network://184.108.40.206 network://220.127.116.11 file:///data # a Mark5 with a linux partition, and also used for eVLBI mark6-01, 1, 0, mark6:// file:///fuse_mark6-01 # a mark6 (native playback) or file-based playback from /fuse_mark6-01
file://<path>data source. In this case the asociated nodes get assigned in a round-robin fashion if
file://<path>data sources appears more than once in the current job.