Need help running your job on the cluster? Need a question answered about a particular tool or cluster resources?
Whenever you have a question, please try to help us help you by providing the following information:
Do NOT use “reply” to an old mail to create a new inquiry. Create a new one, choose a suitable subject line, and do not address it to an individual person, but to the common cluster-help-Adress mentioned above. Otherwise, the topic will be both obsolete and completely different to what you really want, and on top, it will be attached to an old case. You will not get a timely response if we think you are just responding to an introductory course invitation, or if you personally address someone who is currently not at their desk. And your case will get sorted to the bottom of the queue in at least some of the to-do lists we have due to its old age — and thus easily get overlooked.
State your username (nh…), Job ID(s) and locations on the system (directories, hostnames etc.)
NEVER send us your password. We do not need it. Sending your password will lead to your account being locked.
One problem per case: Nobody at the LUIS knows everything. If you pack several different topics into one case, fewer persons are able to answer them all, even if three out of your five questions would be easy to reply to. So the most difficult question in an inquiry will usually influence response time. One question at a time will usually get you answers much quicker. If things could depend on each other, then by all means describe them in one case. Some things are complex.
Details known about the problem, e.g.
JobID(s) of the job(s) that have problems
what command(s) in what sequence lead to the problem? What did you do? In which directory?
Have you checked your quota (use the command checkquota)? (really!)
the time something happened; be precise if you can: log files are big and sometimes there are 1000 entries every minute
the batch script (as an attachment or - even better - its location on the cluster)
output from the program, if available, like myjob.o12345 and myjob.e12345 (as an attachment or - even better - its location on the cluster)
what did you already try to solve the problem yourself?
for batch jobs:
did you request sufficient memory for your job?
in case of problems running a multi-node job: did you already try a single-node job? What was the outcome? Is it even reasonable to run your job over several nodes when it would fit on a single node (intra-node communication between processes may be a lot faster than inter-node communication)?
is the combination of cores, nodes and memory reasonable or even possible on the system? Did you check the
list of nodes available to find your setup, or did you just copy a very old setup for a partition that does not even exist any more from a former colleague of a former colleague?
make sure you do not change/delete the files we need to trace your problem while we are at the case. The best method is to provide a minimal setup in a separate directory that we can use. The simpler and clearer your bug report, the easier it is to isolate the problem. A very good case shows “this runs ok” in comparison to “this does not” (e.g. a smaller job setup, less single/multi nodes, memory request, different parameter, different environment variable…)
for questions related to file transfers or graphical logins:
how do you access the cluster? Weblogin? PuTTY? X2Go? SSH? What is your workstation's
OS?
what is unusual or unexpected in the output files you already have?
bus errors and segmentation violations usually are not a system problem. They occur when a program tries to access memory locations or hardware that does not exist, which may happen e.g. when the software has problems with proper use of pointers or when it does not deal properly with unexpected input data.
Screenshots help, but consider doing copy&paste in plain text for textual information
After you got your response, do not just reuse the case to ask for something completely different. The person who replied to your first case may not know the answer to your new problem, but the case remains assigned to the same agent. One case per problem.
In case you solve your problem yourself in the time it takes us to answer your case, please tell us. Ideally, you also tell us how you solved it: another user may have the same problem one day, and we do not know everything.
We will NOT do your Homework for you. That means we expect you to at least read your own log files and check with the documentation and the
FAQ on this site. We also expect you to do your own thinking.
It is not a problem if you have read your stuff and just do not understand what it means. We know this is complex business and we will try to help. But please try to avoid creating a case that just says “my jobs fail” and clearly shows that you neither read any documentation nor used your own brains.