Site definition file format

Some of the values governing the behavior of the load utilities and daemons will most likely vary from site to site. For example, values naming pathnames to files may differ. These values may be specified in a site file, or specified with environment variables. An environment variable definition overrides a definition in the site file.

For example, when the PCT tries to determine whether the user executable will fit in RAM disk, it needs to know the path to the RAM disk. If the environment variable PCT_SCRATCH is set, the PCT uses this value as the path name. If this environment variable is not set, the PCT looks in the site file for the string PCT_SCRATCH and takes the string following it to be the path name. If there is no site file, or no PCT_SCRATCH entry in the site file, then the PCT uses the default of /tmp/pct-scratch which is compiled into the code.

The site file location is named in the source file top/include/config/cplant.h as /cplant/etc/site. This pathname may be overridden by defining the environment variable SITE_FILE to be the pathname of the site file.

The format of the site file is a list of definitions. A definition is an environment variable name, followed by the value, delimited by white space. Other text can appear before or after this definition. The load utilites seek the environment variable name string in the file, and then take the following string to be it's definition, so don't name the variable elsewhere in text.

Most variables can have the special value none meaning in general that that resource is not provided on your machine.

The following definitions can be made in the site file.

Site file definitions
Name
Default if Not Set	Description
PCT_SCRATCH /tmp/pct-scratch	To start a user process on a diskless compute node, the PCT must first copy the executable to RAM disk. This environment variable defines the path name of the directory wherein user executable file will be written.
PCT_OVERFLOW none	If there is local disk storage on the compute node, or you have some alternative to RAM disk, recompile the PCT with HAVE_WRITABLE_DISK defined and set PCT_OVERFLOW to the path name on the disk to which applications can be written. The PCT will attempt to write the executable file to this location if there is insufficient space in RAM disk. Set to none if there is no such file system.
LOGFILENAME /tmp/userlog	yod logs a Cplant job to a log file when the parallel application completes. The bebopd logs Cplant jobs to another log file with a format identical to that of ASCI Red. The name of the yod log file is defined here. The bebopd log file is written to the same directory, but has the name run.log.mm-dd-yy. Set LOGFILENAME to none if you want to turn off logging.
VM_NAME_FILE /cplant/etc/vmname	Location of a text file containing the name of the virtual machine. If you don't use virtual machine names, set this to "none".
NOTIFYLIST none	yod can send mail to administrators when a load fails detailing the physical node numbers of the compute nodes that are failing. The NOTIFYLIST is the list of recipients of this mail. Set to none if you don't want yod to send mail upon load failures.

Site file definitions
Name
Default if Not Set	Description
BEBOPD_RESTART_FILE /tmp/saved_pct_list	Upon exit the bebopd writes a restart file listing the PCTs in the virtual machine. This is the pathname to which the restart file is written. Typically, /etc/local is a link to global writable storage that is private to the virtual machine.
NODE_NAMES_FILE /cplant/etc/cplant-map	Location of a text file containing a list of compute node names in physical node number order. The names are displayed by yod and pingd.
CPLANT_HOST_FILE /cplant/etc/cplant-host	Location of a text file naming the node on which the bebopd is running.
PARALLEL_FILE_SYSTEM /enfs/tmp	Path name to a global parallel file system. When executables fail to fit in RAM disk, yod will copy them here and the PCTs will exec them from here. Set to none if you have no such file system.
PBS_PREFIX /	The pathname where the PBS bin, sbin, and man directories may be found.
PCT_HEALTH_CHECK zombies:memory:scratch	The PCT can check the health of the compute node every time it receives a status request from the bebopd. (It omits the check however if it is running an application process.) It can check for zombie processes, low free memory, and insufficient RAM disk space. Set this value to the parameters you would like the PCT to check, or set it to none if you don't want the PCT to check these. The PCT will re-read the site file upon receiving a SIGHUP, so you can change the health check parameters while the PCT is running.