UC San Diego SearchMenu

Running big jobs, issues and strategies

ACMS instructional systems are shared by many students. If you run jobs that make heavy use of system and network resources, it is important to consider the potential impact the jobs will have on other students. Use the guidelines in this article to run your jobs in a way that will minimize interference with work being done by other students.

General guidelines:

  • Run big jobs on workstations, not on servers such as ieng9, iacs5, insci14.
  • Go to the lab and sit at the workstation. Or, if you have an account on ieng9, see the information about the "UAPE 200" machines below.
  • If you are not seated at the workstation, consider using "nice" to lower the priority of the job.
  • Be aware that running X clients on ACMS machines with the display directed elsewhere (e.g. to your PC running an X server) can cause excessive network traffic depending on the X client application.
  • If you are not seated at the workstation, run jobs at "off peak" hours to reduce the chances that you will interfere with students working in the lab.
  • Structure jobs using a modular approach so that work is done in stages; intermediate results are captured; and the job does not have to be restarted from the beginning in the event that it is interrupted.

Explanation of guidelines

Big jobs should be run on workstations instead of servers because the server systems cannot support long running compute intensive jobs and remain responsive to primary functions: interactive computing, authentication, file service, etc.

UAPE 200 machines

There are four Sun Solaris compute servers located in the main ACMS computer room which have been designated specifically to run compute intensive jobs for Jacobs School of Engineering students and classes. If you have an account on ieng9, you can run jobs on any of these hosts. Since the machines are in the ACMS Computer room and are not directly accessible you will need to login remotely using ssh. The hostnames are: uape-200.ucsd.edu through uape-203.ucsd.edu.

Running big jobs on workstations confines the computing load to that system. If you are in the lab seated at the console, you may notice that the workstation is slowed by your job, but that will not affect anyone else. On the other hand, if you run a job in the background on a workstation and you are not occupying the seat at the console, another student may try to use the workstation and find it unresponsive. In this situation background jobs that cause complaints will be terminated.

Running a job at a lower priority will reduce its impact in terms of competing for processor time. In general this is good thing to do if you must run a job on a workstation and you will not be occupying the seat at the console. The Unix "nice" command is be used for this purpose. Note however, that jobs can also impact performance by consuming large amounts of RAM memory. In that case, lowering the priority of the job with "nice" has the adverse affect of prolonging the time that the memory is occupied.

Schedule work during times when labs are least likely to be busy, e.g. early mornings hours, and perhaps weekends.

ACMS does not promote or provide technical support for running X clients on ACMS machines with the display directed elsewhere (e.g. to your PC running an X server). This is partly due to the same computing load issue discussed above, but is also due to the fact that the remote X server arrangement has the potential to cause excessive network traffic. For example, some Web pages contain mechanisms that are automatically downloaded to the browser and which then generate frequent updates to the browser window. If you were to run the browser on an ACMS machine and display the browser window on another machine (your PC), the network could be saturated with traffic corresponding to the frequent screen updates. This problem is avoided by running the browser directly on the machine (your PC) with the display.