Condor can be downloaded from the University of Wisconsin website at the address http://www.cs.wisc.edu/condor.
The version installed and used at the CMS TOB lab is 6.6.0 (rel. Nov 2003).

The philosophy behind condor is that you have a pool of machines, one of which is the central manager, and the others are job-running machines.

Detailed information on how to install condor on Linux/Unix machines can be found at the above website, in the Unix v6.6 installation page. From the site home page you will also find information about other platforms. Remember that you can make a pool with machines with different platforms on them.

The choices we did were to have a condor user/group, and follow the relative instructions in the tutorial. We also decided not to rely on a shared filesystem for binaries installations, so in the installation script the answer to that question is no. Every machine has its local installation under /usr/local/condor.

If you want to enable Java support in condor (recommended), make a Java installation somewhere BEFORE installing condor. You will need the path to the JavaVM (i.e. <java inst dir>/bin/java).

To start installation, first of all create a condor user/group, then run as the root user the condor_install script in the directory where you untarred the downloaded package.
Running the script, you can use the default answers to almost every question; just remember to say that:

• you want to do a full installation

• the machines are NOT using a shared file server and do NOT have a common UID domain

• the central manager is pcepcmt19.cern.ch (give the full hostname, you never know)

• the person who should receive mail notifications is (at the moment) paolo.bartalini@cern.ch

Other settings which differ from the standard installation, and that you will need to manually edit after installation, are:

• the timeouts: in the /usr/local/condor/etc/condor_config file, set both ContinueIdleTime and StartIdleTime to 1 * $(MINUTE)

• host permissions: in the same file, set both HOSTALLOW_READ and HOSTALLOW_WRITE to *.cern.ch

• a symbolic link added: make a symlink from /etc/condor/condor_config to /usr/local/codor/etc/condor_config. This should not be necessary, according to the manual, but it seems to be needed.

Also one thing: it is not 100% clear to me as why this happens, but jobs will not run if the /home/control directory and its subdirectories are not open in rwx to all users. This is a bit strange (because jobs should be executed as the condor user), and for sure it is not nice, so if anyone manages to understand this, please feel free to correct. To fix this:

chmod -R go+rwx /home/condor

The last step to do is to set up the machine so that it starts condor at boot time. Copy the /usr/local/condor/sbin/condor_master script to /etc/init.d and create symlinks to it in /etc/rc0.d and /etc/rc5.d. Then reboot your computer, and run condor_status at the command line to see if the machine is now in the pool (it should appear in the list of nodes).

The above instructions apply both for setting up a central manager machine, and for adding a machine to a pool. The difference is that, when you set up the central manager, you will specify that the central manager's name is the name of the machine you are on, and condor will understand what you want to do.