dcsimg

San Diego Supercomputer Center Selects Appro as One of the First Vendors for ‘Triton Resource’

The San Diego Supercomputer Center (SDSC) at the University of California, San Diego has selected the first round of vendors for its new Triton Resource, an integrated, data-intensive computing system primarily designed to support UC San Diego and UC researchers. Triton is scheduled to enter production early this summer.

The San Diego Supercomputer Center (SDSC) at the University of California, San Diego has selected the first round of vendors for its new Triton Resource, an integrated, data-intensive computing system primarily designed to support UC San Diego and UC researchers. Triton is scheduled to enter production early this summer.

Plans for the new shared resource cluster, data analysis facility, and large-scale storage system were first announced last fall as SDSC formally opened a new building and data center that doubled the size of the existing supercomputer center to 160,000 square feet. Triton was designed collaboratively with University researchers and computing specialists at SDSC.

“The Triton Resource has first and foremost been designed to integrate with the analysis workflows of UC researchers,” said Philip Papadopoulos, director of UC Systems at SDSC, and head of the definition and acquisition team for the new system. “The design and evaluation team paid special attention to the movement of data within Triton and ensuring scalable network connectivity to other resources. These companies were selected for their leading edge, energy-efficient systems that met very specific criteria and a unique set of requirements.”

The first phase of the Triton Resource was procured through a highly competitive ‘request for proposal’ process, in which vendors bid for one or more of the system’s three major components: a shared resource cluster, large memory cluster, and a high-performance networking fabric. A scalable, parallel file system will be procured through a similar process later this year. The entire system is integrated using the Rocks Cluster Toolkit, developed by researchers at UC San Diego and supported under National Science Foundation grant #OCI-0721623.

The Triton Resource is named after the mythical sea god and his three-pronged trident whose image was adopted by UC San Diego. The Appro system selected consists of the following:

  • Triton Compute Cluster (TCC): A scalable cluster configured to operate either in a standard batch mode, or be set up so users can run customized software stacks at scale, with full connectivity to large-scale storage. Appro International, Inc., of Milpitas, Calif., and its GreenBladeâ„¢ server series, was selected by SDSC for the cluster’s computing platform, along with Nehalem processors from Intel Corporation, of Santa Clara, Calif. The TCC has an aggregate of 6TB memory across 256 nodes.

These major resource components are interconnected with a:

  • High-Performance, Low-Latency Fabric: The full bisection Myrinet Fabric and Myricom Dual Protocol Network Interfaces from Myricom, Inc., of Arcadia, Calif., were selected to give the Triton Resource excellent scaling for parallel message passing codes and unprecedented integration with campus, national and international 10Gb/s networks. In addition to the 480 Fabric Ports, the core switch includes 32 10Gb/s Ethernet ports, enabling Triton’s nodes to direct high-speed network access to other resources outside SDSC.

Connectivity for the Triton Resource to UC San Diego campus laboratories will be achieved through both production and research (NSF-funded OptIPuter and Quartzite) multi-10 Gigabit networks to allow for unprecedented integration into research laboratories. Connectivity for UC researchers elsewhere will be achieved using a 10 Gigabit Ethernet campus connection to the statewide education network designed and managed by the Corporation for Education Network Initiatives in California (CENIC).

Full details and specifications of SDSC’s Triton Resource will be announced starting in summer 2009.

Comments are closed.