Interactive Apps with Open OnDemand

Open OnDemand’s File Explorer, the FastX Web interface, and various command-line interfaces, can be used to prepare work for the cluster. This includes transferring and editing files, looking at output, and so forth. However, all production work must be run on the compute nodes, not on the frontends.

A large, multi-user system like UVA’s HPC cluster must be managed by some form of resource manager to ensure equitable access for all users. Research Computing uses the Slurm resource manager. Resource managers are also often called queueing systems. Users submit jobs to the queueing system. A process called a scheduler examines the resource requests in each job and assigns a priority. The job then waits in a queue, which Slurm calls a partition, until the requested resource becomes available. A partition is a set of compute nodes with a particular set of resources and limits. There are partitions for single-node jobs, multiple-node jobs, GPU jobs, and some other dedicated partitions. A list of the different queues and resources are listed below.

Open OnDemand offers an easy way to run interactive jobs. With an interactive job, you are logged in directly to a compute node and can work as if it were a frontend. Please keep in mind that an interactive job terminates when the time limit you specify expires, unless you explicitly end the session.