PBS (Portable Batch System, [2]) is an open source system of daemons and utilities that may be used to schedule jobs on a parallel machine or distributed computing resource. We have made several modifications to PBS and some of these are described at [1]. We have also made changes to the Cplant runtime components to integrate them into PBS.
To run Cplant jobs under PBS, users generally write a script that runs yod to launch their application. They submit the script to PBS, and PBS schedules it to run at a later time. When they submit the script they also request a certain number of nodes and the length of time they wish to use the nodes. Their job script can run yod many times, in succession or in parallel. But at no time may their job script use more total nodes than they requested at submission time. Complete instructions for using PBS are given at [1].
This chapter describes the PBS components, and the enhancements to Cplant that support PBS. It also describes the PBS configuration and log files that are of use in configuring and trouble shooting PBS. Some common runtime problems are described as well.
The details of running PBS are described in the PBS Administrator's Guide and in the PBS External Reference Specification, both available at the PBS web site [2]. (The Administrator's Guide can be downloaded directly. The ERS must be built from the PBS source tree available at [2].)