FAQ

What queue should I use? The answer depends on number of processors needed and memory use. See Queues for information on Saguaro's queues.

What compiler should I use? Each compiler will optimize your program differently. Sometimes a program will compile with one compiler, but not with another. In this case use the compiler that compiles your code. If all the compilers produce functional programs the choice comes down to which compiler produces the fastest running program.

What memory limitations do I have?1 GB/Processor. If you need more, please submit a service request.

I have a file that looks correct, but it won't run! If the file was created or modified anywhere other than Saguaro, run dos2unix filename.sh. This may fix the problem. If this does not fix the problem, please log a service request. Reason this works: On non-Unix systems (Windows for example), the end of every line contains 2 characters- a carriage return and a line feed (\r\n). Unix expects the end of a line to only have a line feed (\n). Having the carriage return confuses the scheduler, and prevents your job from executing. Running the dos2unix command converts the problematic characters into characters that Unix can deal with.

My job output has errors like this: /usr/X11R6/bin/xauth: error in locking authority file This error should not affect the job at all. This error indicates that when you logged into Saguaro with SSH, X-forwarding was enabled. The job scheduler does not care about X-forwarding since it does not affect the application. This error can be stopped by changing the SSH client login options to disable X-forwarding or tunneling. If for some reason the application does crash and this is the only error, put in a service request.

What is checkpointing? When a job is checkpointed it will periodically write its state to disk. The job can then use this information to start where it left off (well, really where it was last checkpointed). Checkpointing allows your job to overcome problems that cause the job to stop before it completes, like running out of your estimated wall time, running out of disk space, or a hardware failure. Checkpointing is already supported in many scientific applications, you just need to turn it on!

How do I estimate walltime? Run a small form of your job in the devel queue, then using that time information extrapolate to the larger number of processors and job size. Then add a buffer for when the job runs longer than estimated.

I cannot write to my home space anymore, it says the disk is full, but df says there is still free space? What is going on? You are probably over your quota. Please read the section on tracking your disk usage for more details.