Short Course: Setting up and Using a Small Linux Cluster

July 8, 2001, Town and Country Resort Hotel, San Diego, California

Organizer:

David Moulton, Los Alamos National Laboratory

Description:

The primary objective of this workshop is to communicate the necessary information about hardware, software and application considerations so that participants can analyze their own needs and potential use of clusters. To this end we will provide an overview of hardware related issues, a comprehensive list of freely available software resources (both administrative and developer related), and an overview of the computational work performed on the Avalon cluster (http://cnls.lanl.gov/avalon) at the Los Alamos National Laboratory. To help make this truly practical we will bring a small cluster and demonstrate various stages of construction and configuration as well as compiling and running a parallel code.

Level of Presentation:

30% Introductory, 40% Intermediate, 20% Advanced

Who Should Attend?

This short-course is intended for faculty, students, professional scientists and engineers that are interested in parallel computing but generally find the necessary resources are either too costly or simply unavailable. Specifically, we'll demonstrate the potential of low-cost and reliable small to medium size Linux clusters that are suitable for research groups and departments.

About the Instructors:

The instructors are but a few of the many scientists in the Theoretical Division at Los Alamos National Laboratory (LANL) that use Linux clusters in their research. Additionally, they are representative of a smaller group of scientists that develop and support Linux clusters within the Theoretical Division, which range in size from workgroup (i.e., 8-32 nodes) to a division cluster (i.e., Avalon has 140 nodes).

David Moulton
Mathematical Modeling and Analysis Group

Patrick Goda
Mathematical Modeling and Analysis / Theoretical Astrophysics Groups

Michael Warren
Theoretical Astrophysics Group

Markus Berndt
Mathematical Modeling and Analysis Group

Outline:

The total time will be approximately. 8 hours, with a break after each section. Each section will be approximately 1.5-2 hours.

Hardware Selection:

• discuss the basic design concept of Beowulf (commodity) clusters
• overview of the hardware components (e.g., processor, memory, network, hard disks )
• price/performance trade offs and application customization

Assembly and Configuration:

• issues related to prototyping, ordering parts/systems
• space, power and cooling considerations - automated installation and maintenance of the nodes ( e.g., groups such as OSCAR with LUI, and ROCKS with Kickstart )
• batch processing: do you really need it?

Software Tools:

• compilers and debuggers (vendor, GNU) (myth: just use the -parallel flag)
• message passing interface (MPI) (various incarnations and performance differences)
• utilities, tools and useful libraries

Some REAL Example Codes:

• an overview of 3 years of research on Avalon, the 140 node Alpha/Linux cluster in the Theoretical Division at LANL.
• discuss on our own work with parallel multilevel elliptic solvers
• highlight some commercial codes that run on clusters

Additional Information:

For additional information about this workshop and cluster related links check out our web site at http://math.lanl.gov/SIAM01-ClusterWorkshop/.

Registration:

Seats are limited. Please register before the deadline. To register, please submit the Preregistration Form. Submit completed form with registration payment to reach SIAM on or before June 7, 2001. Registration fee includes coffee breaks and lunch on Sunday, July 8.