CS290I Lecture notes -- Globus in 15 Minutes

• Rich Wolski
• Globus fast: http://www.cs.ucsb.edu/~rich/class/cs290I-grid/Globus/globus-ucsb.html
• Useful Globus References:
  • The Globus Home Page
  • The Globus Documentation Page
  • The Globus Quick-start guide (35 pages)
  • The globusrun man page
  • The GRAM Page
  • The RSL Page
  • The Globus API Reference for Release 2.0
• The Globus lecture from class.
• The NWS lecture from class.
• The NWS LDAP lecture from class.

  ---

  This tableau is intended to get you started with Globus in something less that the 35 pages contained in the Globus Quick-start guide. These instructions are valid on any machine listed in the machine list for this class, and no others. That is not to say that some variation will not work in other environments, but these examples work exactly as depicted on the machines in the machine list.

  ---

  WARNING: Graziano Obertelli and Martin Swany are NOT TAs for this class, nor are they your tutors. They are professional research staff members of my group who have installed Globus and the NWS (respectively) on the UCSB instructional machines as a courtesy. DO NOT contact them directly with questions or bug reports. That includes email contact, phone contact, and personal contact. There are ample Globus and NWS references and documentation, many pointers to which are shown above. Graziano and Martin are not reference material nor are they on-call to help you use Globus or the NWS. All questions regarding Globus and the NWS must be directed to The TAs at all times (including times when you think these systems aren't working and the TAs are not immediately available). Your grade in this class will suffer severely if you do not direct all questions regarding Globus and the NWS to either The TAs or to me. The only exception to this iron-clad requirement is when you email Graziano your user-cert-request as described below.

  ---

  Setting up your Globus Environment

  If you have never had the pleasure of Globusing before, it is fun, but you have to get things initialized properly to take part in it. Here are the steps you need to take to set your user account up for the Total Globus Experience. You need to execute these in the order they are listed.
  1. You must add to your environment the following variable initializations:
     • for bash or sh:

       			export GLOBUS_LOCATION=/usr/local/globus
       			. $GLOBUS_LOCATION/etc/globus-user-env.sh
       		

     • for csh:

       			setenv GLOBUS_LOCATION /usr/local/globus
       			source $GLOBUS_LOCATION/etc/globus-user-env.csh
       		

     These variables tell your shell (and any shells that Globus launches for you) where to find things. It also adds a bunch of stuff to your user environment that you need. If you are like me, you will want to put these in your .bashrc or .cshrc file just for good measure.
  2. You must create, in your home directory, a .globus subdirectory. To do so, type

     		cd ~
     		mkdir .globus
     	

  3. Type the command

     		grid-cert-request
     	

     and follow the instructions. You will be asked to type in a pass phrase (twice). If you lose this pass phrase, you will have to re-do the entire process so please keep track of it. You will then be asked to mail your certificate to the CA (who happens to be Graziano, in our case).
  4. Email the file usercert_request.pem that has been created in your .globus directory to graziano@cs.ucsb.edu exactly as specified in the instructions. Warning: The standard Globus installation will ask you to mail your certificate request to another email address (not Graziano's). We have hacked the local installation to consider Graziano as our CA. If you are not instructed to email your certificate request to Graziano, or if another name appears, something has gone wrong. Make sure you are logged into a machine appearing in the instructional machine list. If you are, email the TAs and let them know.
  5. You will receive a reply from Graziano (quickly, the first time, but more slowly if you make additional requests due to the loss of your pass phrase or other self-inflicted disasters) in which your certificate is contained. You should save the contents of his message in your .globus directory under the name usercert.pem. You can even leave the mail headers in place as Globus is smart enough to find the actual certificate buried in the message.
  And that's it. Your account is now Globus enabled. You only need to do this initialization once (successfully). This next step you need to do every time you log in.

  ---

  Establishing a Globus Proxy

  Before you can run anything (either just after completing the task above or after you have logged in) you need to create what is called a "proxy" for yourself on the local machine. From the command line you type

  	grid-proxy-init
  

  and you will asked for your pass phrase. Once you have successfully authenticated yourself to the proxy, it will handle your authentication duties for a fixed amount of time (24 hours, by default). Every time you invoke a Globus command that requires authentication, the proxy will handle the necessary certificate and RSA challenge functions for you without asking you for a password. Typing

  	grid-proxy-init -timeleft
  

  will tell you how much time is left before the proxy expires.

  ---

  Querying the MDS

  To get a global query of all host information in the MDS, type

  	grid-info-search -x -b 'mds-vo-name=mayhem, o=Grid' -h pompone.cs.ucsb.edu -p 2135 'Mds-Host-hn=*'
  

  all on one line, exactly as shown. If you want information on one host (marge.cs.ucsb.edu, say) type

  	grid-info-search -x -b 'mds-vo-name=mayhem, o=Grid' -h pompone.cs.ucsb.edu -p 2135 'Mds-Host-hn=marge'
  
   

  again, exactly as shown. If you add the .cs.ucsb.edu to name dagwood, it will fail.

  ---

  Querying the NWS with the Globus Tools

  To get NWS data, you have to use the standard LDAP searching mechanism. The reason is that the version of Globus we have installed used a version of LDAP that is somewhat old. grid-info-search is really an alias to an old ldapsearch that is incompatible with our current NWS implementation. Don't ask.

  Anyway, you need to specify a different part for the tree and a different GIIS to use. Getting the NWS CPU data for marge, you would type

  
  ldapsearch -x -h dreadnought.cs.ucsb.edu -p 3389 -b "o=data,service=nws,o=grid" "forecast=*marge*CPU*"
  

  ---

  Running a Job

  To run a job on a Globus-enabled host, you use the globusrun command. The general syntax is

  globusrun -r resource_contact_string RSL_string
  

  In our environment, for reasons explained in the lecture, the contact string contains redundant information. If you wish to run "/bin/date" on marge.cs.ucsb.edu, here is the command you must issue:

  globusrun -o -r marge:2119:/O=Grid/O=ucsb/CN=marge.cs.ucsb.edu '&(executable=/bin/date)'
  

  exactly as shown. If you inadvertently move the ".cs.ucsb.edu" parts of this string around, it will fail.