0-2. Login node and Computation nodes
Most of the parallel computing systems are equipped with computing nodes and login node. Only the computing nodes have system resources to conduct massive computing tasks.
All resources (CPU, Memory, GPU, etc.) for large-scale calculation are installed in the computation nodes. Basically, you can not log in directly, you need to access from the login node via job scripts.
Normally you will not log in directly to a computation node when doing calculations.
Be sure to log in to the login node and register a job script in the queue from the appropriate login node.
Registered job scripts are sequentially executed on the computation node according to the submission order and priority, and the calculation results are returned.
There are no resources (CPU, Memory, GPU etc) available for calculation on the login node.
Performing calculations accidentally at the login node is troublesome to other users.
For this reason, processes executed for a long time at the login node are forcibly killed with the following rule.
Killing processing from the login nodes
1. Processes running more than 2 hours on login nodes are killed and the owener is emailed．Processes in this permitted list are excluded.
2. Time count is CPU time. (In the case a process does not consume much resources, it may not be detected)3. Trial period until 2016 / end of October. During the trial, the actual process killing is performed manually after sending the automatic mail, some time gap until the process is killed.4. Target machines at 2016 - March is the PC cluster (hpcc)
Visualization and Data processing