Parallel performance of the new INGV-PI Laki cluster measured with HPL, HPCG, OpenFOAM and ASHEE

Abstract

Laki is a High Performance Computing system installed at INGV-Pisa to support the fluid-dynamic and volcanology modelling activity. Laki is built with a traditional architecture used for high-performance computing clusters as pioneered by the Network of Workstations project [Anderson et al., 1995] and popularized by the Beowulf project [Sterling et al., 2005]. This system is composed of standard high-density servers, a Gigabit Ethernet network and a high-performance InfiniBand interconnection (Figure 1). We have defined the cluster architecture to contain a minimal set of high-density components in an effort to build reliable systems by reducing the component count and by using components with large mean-time-before-failure specifications. Information and access to specific monitoring tools can be found at http://laki.pi.ingv.it; the site can be reached only from inside INGV network, or through an INGV VPN connection.