wiki:Docs/CAP9.0InstallationArcher

CAP9.0 Installation on ARCHER

On PUMA, the code was extracted from the Met Office Science Repository, using the command,

fcm co fcm:ancil.x-tr -r 307

The revision number corresponds to CAP9.0.

The following three files were created in the fcm-make directory,

  • ancil_ARCHER_cce_debug.cfg
  • ancil_ARCHER_cce_opt.cfg
  • inc/ARCHER-cce.cfg

These were based on the Met Office XC40 equivalents. The changes related to directories on the local computer. The configuration files seem to impose the use of absolute directories.

With these changes, the command,

fcm make -f $HOME/CAP9.0/fcm-make/ancil_ARCHER_cce_opt.cfg --new
> CAP_build.log 2>&1

was used to mirror the code over to ARCHER.

On ARCHER,

cd $DATADIR/CAP9.0
fcm make --new

This built the code successfully. A new branch was created in the Met Office Science Repository to contain the ARCHER variant.

Issues Running the code on ARCHER

Script modification

The scripts AncilScr_name were modified, where name is one of: top, mask, vegfracsoil, ozone, smow, sstice, slt, aerosol, orog.

These were modified by eliminating the decisions based on the HARDWARE variable. The decisions that would have been made on the Met Office XCE computer were selected. Also, the first three were modified for the Cray assign problem.

PBS commands

The PBS commands for XC40 are not compatible with XC30. The ncpus=1 construct is not accepted in the XC30 PBS select command. This led to failure to launch and no helpful error messages were produced. The issues are the specification of the queue is required on the XCE but not on ARCHER.

Broken Links in the Ancillary Tree

The test programs use XML scripts which extract data from the ancillary tree. For example, in the directory ancil/data/parameters/IGBP_to_MOSES_accurate there are links such as,

latest -> /projects/um1/ancil/data/parameters/IGBP_to_MOSES_accurate/v2

where v2 is in the same directory as latest. These links pointing to absolute paths are present on the Met Office computer. I have modified the few on ARCHER necessary to complete the tests but there are 759 in all. There seems to be no reason for this.

Comparison of Computers

The CAP scripts contain the line,

HARDWARE=$(uname -a)

which is then used to determine subsequent actions. The following code snippet from the scripts was run on Met Office computers and ARCHER, producing the results in the table below.

#!/bin/ksh
HARDWARE=$(uname -a)

if [[ $HARDWARE = +(*xc*) ]]
then
  echo Hello
else
 echo Goodbye
fi

if [[ $HARDWARE = +(*Linux*) ]]
then
  echo Hello
else
 echo Goodbye
fi
Met O PC Met O XCE MONSooN (Met O XCM) ARCHER
Goodbye Hello Hello Goodbye
Hello Hello Hello Hello

showing that the Met Office Crays are set up differently from ARCHER.

Last modified 23 months ago Last modified on 09/02/16 13:10:27