ipp user

Operations are generally performed as the ipp user on the Production Cluster. You should add your ssh public key to ~ipp/.ssh/authorized_keys for ease of access.


Pantasks runs in the server/client mode, so that it can easily be accessed by multiple people.

To distribute the load, we use the following for the server:

  • summit copy: ippc01
  • registration: ippc02
  • standard science: ippc16
  • publishing: ippc08
  • distribution: ippc15
  • cleanup: ippc07
  • replication: ippdb00
  • update: ippc13
  • pstamp: ippdb02
  • detrend: ippc06
  • addstar: ipp004

The client can be run on any machine, and by any user, so long as you're in the appropriate directory (which contain a ptolemy.rc file that selects the correct server to contact).

Server Scripts

The input scripts used by the different pantasks servers are maintained in svn under 'operations'. Like 'ippMonitor', the installation of this directory requires site specific information:

% cd ~ipp/ipp-svn/operations
% configure --opsdir ~ipp
% make install

Pantasks Servers

All the servers start exactly the same way. Their input files come pre-configured with a set of hosts, labels (and in the case of stdscience, survey tasks). Everything is ready to go. This is how to start one of the pantasks, for example, SummitCopy:

  • Check which machine to use for the server, above.
  • Start the server:
    % ssh ipp@ippc01
    % cd summitcopy
    % pantasks_server &
  • Start the client:
    % cd ~ipp/summitcopy
    % pantasks_client
    pantasks: server input input
    pantasks: setup
    pantasks: run

If all of the pantasks need to be started or stopped, it is more efficient to use

IPP User backup and restore

Important files in the home directory of the IPP user on the MHPCC cluster are saved to a backup with the script ~ipp/make.backup. This uses rsync to save copies to a target backup location (currently /data/ipp004.0/ipp/operations). To restore from this directory, use rsync.

Clean Checkout and Compile of IPP software

Make sure the trunk compiles. When it compiles, make a tag, for example:

svn copy

and checkout a copy of this tag into ~ipp/src/ipp-20100601

compile it, using:

psbuild -ops -extbuild 

This builds it with the required magic libraries (and errors out if magic cannot be built), and sets the '-dev' option as well (builds ippdb.src rather than using the checked in version).

Note: There is some problem with the build for magic. It fails the first time through. To complete the build use:

psbuild -ops -start magic

If database changes are required, update BOTH gpc1 and isp, to keep them in sync.

Currently, the following files need to be updated, to point to the current build:

  • .tcshrc
  • .ptolemyrc

and the following file will need to be copied to the new location:

  • psconfig/(OLD_BUILD).lin64/share/pantasks/modules/

Check ~ipp/README to see if there are other instructions.

The addstar server is now being run by the ippdvo user, and therefore is not running under the same ipp build. Talk to Addstar Czar (or Heather) for more details about configuring the ippdvo user.

Updating Magic Code

The magic and related software is stored in ~ipp/magic_tarballs with names like magic.20091023.tgz and ssa-core-cpp.20091023.tgz. The active release is linked to ~ipp/magic.tgz and ~ipp/ssa-core-cpp.tgz. If a magic update is provided, place the source code in magic_tarballs with an equivalent name and update the ~ipp link. The new code can be built manually (as opposed to using psbuild -ops), these are the steps. To build the new code, it is necessary to update the tree in the ipp build directory. Go to the top-level magic directory in the tree (eg, ~ipp/ipp-svn/magic) and do the following steps:

make update
make all
make install

The 'make update' step is the critical one here: it untars the code from ~ipp/magic.tgz and ~ipp/ssa-core-cpp.tgz. If the code is present, magic will be build by the standard IPP build system.