Introduction to HTC at CHTC: Glossary

High throughput computing is good for computations that can be broken into individual pieces.

CHTC users can access the CHTC pool, UW Grid and Open Science Grid.

A job consists of a computational task, usually defined by input data and a piece of software, producing output data.

Most large scale systems consist of a head node for logging in and submitting jobs, where jobs are performed on worker nodes.

A batch scheduler controls where and when jobs run on the worker nodes.

A submit file tells the job scheduler information like resource requirements, software and data requirements, and what commands to run.

Jobs may go through various states in the queue, and can be monitored by a queue command.

Successful jobs leave the queue, bring back output files, and have no errors.

Log files tell when and where a job waited and ran, and how many resources it used.

Output files capture any general information printed by the job’s executable.

Error files capture any errors that printed when the job executable ran.

Always run a test job before submitting a full scale job.

To test a new job, use an interactive session beore submitting.

You can use log, standard output, and standard error information to determine why jobs fail.

It is important to follow your resource’s policies and procedures.

Have a data management plan in place before you start computing.