RHIT Parallel Computing - Accounts

Cluster Accounts

Types of Accounts

• Faculty/Staff Account - a permanent account for any type of computing project related to teaching or research.
• Course account for students - for students in course, requested by a course instructor, accounts are removed two weeks after the end of the course; longer if specified by the faculty member. 
• Project account for a student - for a student to work on a project under the supervision of a faculty or staff member, e.g., thesis, project, undergraduate research longer than a quarter. 

Requesting an Account

• Faculty/Staff Account - apply at the helpdesk; or print out, complete, and submit a cluster account request form. Please pay attention to:
  
  
  • If you are going to use the cluster for unusual amounts of time, e.g., 100% of the cluster for one week at a time, or many gigabytes of intermediate data, put that in the special requirements section of the form.
  • If the computing project is a for-fee consulting project also put that in the project description section. For-fee projects need to be approved by VP of IAIT, and fees may need to be charged if the computing is extensive. See account policies below. 
    
    
• Course account for students- apply at the helpdesk; or print out, complete, and submit a cluster account request form. In filling in the account request form, please ensure that:
  
  
• specify the maximum amount of resources to be used by an individual student per job, e.g., a maximum of four nodes for 1 hour, or whatever is appropriate. These maximums can be assigned to the individual student account. The instructor will be responsible to ensure that students understand the special care required in using the cluster resources responsibly. 
• specify the duration of the students accounts. Normally, the accounts will be killed two weeks after the end of the course. For a student account spanning several quarters an individual student project account should be used. 
  
  
• Project account for a student  - apply at the helpdesk; or print out, complete, and submit a cluster account request form. Please ensure that:
  
  
  • In the special requirements section specify the maximum amount of resources to be used  per job, e.g., a maximum of eight nodes for 4 hours, or whatever is appropriate.
  • The faculty sponsor must sign the request. The faculty/staff sponsor will be responsible to ensure that the students understand the special care required in using the cluster resources responsibly.
    
    

Account Policies

• General
  
• Users need to be especially sensitive that their job may hog resources, perhaps unknowingly. Users will have to be proactive about cooperating.
• Faculty/Staff accounts will have no limit on resources.
• Student accounts need to have limits set by course instructors or faculty sponsors.
• Course instructors and student advisors need to ensure that their students understand the policies, and to set resource limits if necessary. 
• Frequent email communication with other users can help to informally schedule long jobs. 

• Technical
• Generally we should avoid using the head node (called master) for heavy computing tasks. This node is used for all sorts of administrative stuff such as logins, ssh and sftp sessions, running editors and compilers, saving data to home directories, and so on. Thus, if we use the head node for heavy computation along with the other nodes the processing of administrative tasks could get bogged down, and logins and compiles could go very slowly. It is reasonable to use the head node to direct computational tasks of the other nodes. It is fairly easy to write MPI programs that avoids using the head node for heavy computation by using the -nolocal option for the mpirun command. (see the commands.pdf file on the activities page)
• Data produced by computations will be stored on the hard disk of the head node, as there is a performance hit if the data is stored to the user's AFS account. This should only be temporary storage. Users should transfer their data to their AFS account or desktop/laptop. 
• Long running applications - those processing days or weeks - should have a mechanism for saving data to disk and restarting the computation. Thus the calculation could be performed in off-peak hours, over a series of days.

Open MP Accounts (AFS)

Account Policies