HPC/Support: Difference between revisions

From CNM Wiki
< HPC
Jump to navigation Jump to search
Line 55: Line 55:
We may ask you to prepare a ''test directory'', which should help to:
We may ask you to prepare a ''test directory'', which should help to:
* Reproduce your problem.
* Reproduce your problem.
* Diagnose your problem, especially subtle or rare errors that are difficult to reproduce, such as where your application runs but does not fail outright.
* Diagnose your problem, especially when errors are subtle or difficult to reproduce, such as where your application runs but does not fail outright.
* Confirm resolution.
* Confirm resolution.



Revision as of 20:23, June 21, 2021

Review the FAQ list

For frequently asked questions regarding the Carbon HPC system, review Frequently Asked Questions to find answers.

Compose a support request for CNM HPC

If your question is not answered in the FAQ:

  • Submit a support request – Log in with your regular Argonne username and password (same as on Carbon).
  • Choose a suitable Request Type and provide the details as shown in the next section.

Required information

Be as specific as possible in your free-form description or form input fields.

For all issues

Provide the following:

  • The command you were trying to run or the menu item you chose in a graphical user interface
  • The exact error message you get. Copy & Paste the message text, or take a screenshot and attach it to your mail.
  • The directory path and file name you were working on, if applicable.
    It is generally not necessary to include in your message copies of files that already reside on Carbon.
  • Fill in an appropriate email Subject or form Title.
    Best do this after you typed and reviewed the main text of your request.

For issues with PBS jobs

In addition to the preceding section, for questions concerning PBS jobs, provide:

  • the working directory of the job
  • the name of the job file used
  • the job number (for queued or running jobs)

For remote access

When you cannot log in to mega or a Carbon login node, provide:

plus the following:

  • The hostname you ran the command on. It is often shown in the command prompt. (If the prompt shows localhost it is not specific enough – ask your local support person for help.)
  • The hostname you're trying to reach.
  • The username you use to connect.
  • The software name and version you use to connect (e.g. SSH, VNC, or a browser).
  • Your operating system and version.

If you use ssh or scp, repeat your command in verbose mode by supplying the "-v" option twice:

scp -v -v [… other options and arguments …]
ssh -v -v [… other options and arguments …]
Copy and Paste the possibly multi-page output into your request.


Test directory

We may ask you to prepare a test directory, which should help to:

  • Reproduce your problem.
  • Diagnose your problem, especially when errors are subtle or difficult to reproduce, such as where your application runs but does not fail outright.
  • Confirm resolution.

Prepare the test directory so that it contains only files to run your application, either interactively or in a single PBS job, and possibly a file with correct output.

Preferred location
  • In or under a sibling directory of one of a your failed runs (as opposed to a child directory).
Include
  • PBS job file, if applicable,
  • Run-specific data and script files (files named in input data or hardcoded in the code, possibly as symbolic links),
  • If available, a file with the expected (correct) output, but with a file name different from the one generated by the job, such as by adding .ref.

In the course of diagnostics, we will suitably duplicate your directory as needed.

Receiving support

When you get our support response:

  • Carefully read it.
  • Follow all instructions and answer all questions.
  • For account- and password-related issues: it may take several hours for changes to take effect. If your initial attempt fails, wait at least that long before retrying.

Additional considerations

To help us diagnose a problem: