Book Image

Distributed Computing with Python

Book Image

Distributed Computing with Python

Overview of this book

CPU-intensive data processing tasks have become crucial considering the complexity of the various big data applications that are used today. Reducing the CPU utilization per process is very important to improve the overall speed of applications. This book will teach you how to perform parallel execution of computations by distributing them across multiple processors in a single machine, thus improving the overall performance of a big data processing task. We will cover synchronous and asynchronous models, shared memory and file systems, communication between various processes, synchronization, and more.
Table of Contents (15 chapters)
Distributed Computing with Python
Credits
About the Author
About the Reviewer
www.PacktPub.com
Preface
Index

Running a Python job using PBS


The Portable Batch System (PBS) was developed for NASA in the beginning of the 90s. It now exists in three variants: OpenPBS, Torque, and PBS Pro. These are all forks of the original codebase, and have a very similar look and feel from the user perspective.

Here, we will look at PBS Pro (which is a commercial product by Altair Engineering and is available at http://www.pbsworks.com), but most of the features and directives we will mention should work on Torque and OpenPBS with just some minor differences. Also, in the interest of brevity, we will mostly concentrate on the differences between HTCondor (described earlier) and PBS.

Conceptually, PBS is just like HTCondor, a resource manager and job scheduling system. Both have a similar architecture, with a master node (pbs_server), a negotiator and scheduler (pbs_sched), and job supervisors (pbs_mom) on the execution nodes.

Users submit jobs to a queue. Usually, there are multiple job queues for different types...