skip to main content
MGRID - Creating a Shared Cyberinfrastructure for the University of Michigan
Related Links
home > news Back to News Table

MGRID Seminar on Rocks Cluster Toolkit
Presented by: Federico D. Sacerdoti
San Diego Supercomputer Center, San Diego

Thursday, April 28, 2005, 1:30PM at 1180 Duderstadt Center


Commodity clusters have established themselves as the machine of choice for high performance applications. The Beowulf method of building these parallel machines was proposed by Thomas Sterling and Donald Becker in 1993, and fast became the dominant configuration for Linux clusters: commodity components connected by high speed Ethernet or Myrinet-class interconnects. Practice has shown that while the hardware for clusters is cheap, easily available, and well supported by Linux, the process of installing and maintaining the Operating System and essential middleware (batch and grid systems, interconnect support) on N tightly coupled machines is a challenge.

The free, NSF-sponsored Rocks Cluster Distribution has emerged in recent years as the premier method of building and managing the software stack on clusters up to 1000 nodes. Rocks employs a complexity-hiding philosophy that installs and configures nodes with minimum human interaction, relying instead on a priori codification of installation procedures that can be replicated reliably and efficiently. This talk will present the Rocks system, its purpose, structures for installation, monitoring and security, and the Roll extension mechanism. The largest known production Rocks systems will be discussed, along with competing toolkits.

Federico D. Sacerdoti is currently a member of the Cluster Development group, San Diego Supercomputer Center (SDSC). Mr. Sacerdoti has been a core designer for the Rocks cluster distribution effort at SDSC almost since its inception in 2000. Additionally, he has contributed to influential HPC projects including the Ganglia monitoring system and the KeLP parallel message-passing libraries. Mr. Sacerdoti received his BS in Computer Engineering from Washington University in St. Louis,
and MS in Computer Science from University of California, San Diego.
His thesis work was in the area of dynamic cache optimizations for parallel applications.

Directions and Maps to the Pierpont Commons can be found at:


The MGRID initiative is intended to create a collaborative research and development center for faculty, staff and students from participating units, with a central core of technical staff, led by a nationally recognized leader in grid computing. The goal of the MGRID initiative is to develop ... read more arrow

Regents of the University of Michigan Sitemap Contact Us Accessibility Features