Building GraphBased Symmetric Cluster
Felix P. Muga II
Mathematics Department, Ateneo de Manila University,
Quezon City, Philippines
fpmuga@admu.edu.ph
Rafael P. Salda~na
High Performance Computing and Networking Laboratory,
Ateneo de Manila University, Quezon City, Philippines
raf@admu.edu.ph
William Emmanuel S. Yu
High Performance Computing and Networking Laboratory,
Ateneo de Manila University, Quezon City, Philippines
wyy@admu.edu.ph
ABSTRACT -- In August 2000 the High Performance Computing Research Group (AHPC) of the Ateneo de Manila University built an 8-node Beowulf- class computer designed for computational science applications. As more researchers and students in the University are trained in cluster computing, the need for building a better cluster arises. This year the AHPC proposes to build a large-scale graph-based symmetric cluster. The proposed high performance computing system will be a symmetric cluster with a single-switch latency and f lat networking neighborhood topology. The proposed design also features minimized cost and maximized bandwidth. This presentation will deal with mathematical and computational aspects of graph-based clusters, and design considerations for a large-scale symmetric cluster with a single-switch latency.
KEYWORDS -- symmetric, balanced and flat network neighborhoods, isomorphic graphs, bisection bandwidth, pairwise node bandwidth, regular graphs, switch latency, cluster computing, beowulf, parallel computing
|
|
|