Xgrid Facility at CAL

Contact: Kristen Menou, Cameron Hummels, Dave Spiegel

A new computing facility is now available at CAL: an Xgrid, i.e a cluster of Macs with Ethernet interface between nodes. CAL members can start using CAL-Xgrid and contribute to the grid by offering their computer idle time.

Here is how it works in a nutshell. You have an Intel Mac or get an account on callattice, you compile your code, you submit to CAL-Xgrid and you retrieve your results once completed.

Limitations:

Only people with Macs can send executables directly to the grid, otherwise you must get an account on callattice.astro.columbia.edu in order to compile and submit your code to the grid (talk to Kristen). This is not a replacement for parallel calculations on beehive, in that there is no MPI support. It is useful mostly if you have several serial (as opposed to parallel) jobs to run on separate processors, i.e. "brute force" parallelization. Because the grid nodes are Intel Macs, currently only executables produced on Intel Macs can be run (but see below).

The grid nodes are 8 Imacs with 2x2GHz processors and 2GB of memory, currently sitting in the 14th floor library. In addition, Peter Trip from the Psychology department has made available to us 20 Imacs with 2x2.2Ghz processors. The agents accept jobs when they are idle.

Use:

For security reasons, we are not providing the Xgrid controller name and access password (written ?h? and ?p? in the commands below). If you want to use the grid, ask Kristen directly.

  • Submit a job (a.out):
    xgrid -h ?h? -p ?p? -job submit -email ?you?@astro.columbia.edu a.out
    

Note: ?you?@astro.columbia.edu should be your email address. It will tell you when you job starts, finishes and is a record of the "job_id" number (used by the next command)

  • Once the job is done(email notification), retrieve your results (standard output and output files):
    xgrid -h ?h? -p ?p? -job results -id  job_id  -out ./
    

The last two arguments tell Xgrid to return job files in the current directory (./)

The more efficient and reliable way to submit jobs to the Xgrid, however, is to use "Grid Stuffer". You can find information here: http://cmgm.stanford.edu/~cparnot/xgrid-stanford/html/goodies/GridStuffer-info.html

Further Information:

  • Compilers: g77 (from hpc.sourceforge.net, Oct 06 version) has been installed on the nodes and tested with Xgrid. (g95 and gfortran will also be setup at some point in the future). Xcode Tools has been installed as well, so all the c compilers should be loaded properly. If there is a library or compiler you want on the grid, please request it from Kristen or Cameron.
  • For more info on various Xgrid user options: "man xgrid"