JuRoPA System Configuration


System Architecture

Network Topology

Switch Topology


JuRoPA Components

  • Juropa-JSC:
    FZJ/GCS production system
    JUMP (IBM Power4 compute cluster) successor
  • HPC-FF:
    High Performance Computing For Fusion
    Dedicated to European Fusion Research Community

Hardware Characteristics

Juropa-JSC

  • 2208 compute nodes:
    • 2 Intel Xeon X5570 (Nehalem-EP) quad-core processors
      • 2.93 GHz
      • SMT (Simultaneous Multithreading)
    • 24 GB memory (DDR3, 1066 MHz)
    • IB QDR HCA (via QNEM Network Express Module)
  • 17664 cores total
  • 207 Teraflops peak performance
  • 183.5 Teraflops Linpack performance
  • Sun Blade 6048 system
  • Infiniband QDR with non-blocking Fat Tree topology
  • Sun Data Center Switch 648

HPC-FF

  • 1080 compute nodes
    • 2 Intel Xeon X5570 (Nehalem-EP) quad-core processors
      • 2.93 GHz
      • SMT (Simultaneous Multithreading)
    • 24 GB memory (DDR3, 1066 MHz)
    • Infiniband Mellanox ConnectX QDR HCA
  • 8640 cores total
  • 101 Teraflops peak performance
  • 87.3 Teraflops Linpack performance
  • Bull NovaScale R422-E2 technology
  • Infiniband QDR with non-blocking Fat Tree topology
  • Mellanox MTS3600 Switch

Complete System

  • 3288 compute nodes
  • 79 TB main memory
  • 26304 cores
  • 308 Teraflops peak performance
  • 274.8 Teraflops Linpack performance (TOP500 list June 2009)

Lustre Storage Pool

  • 2 Meta Data Servers (MDS)
    • Bull NovaScale R423-E2 (2 x Nehalem-EP quad-core)
    • 23 TB for meta data JSC (EMC CX4-240)
    • 17 TB for meta data FF (EMC CX4-240)
  • 14 Object Storage Servers (OSS) for JSC
    • Sun Fire X4170 Server
    • 500 TB user data (28 x Sun Storage J4400 Array)
  • 8 Object Storage Servers (OSS) for FF
    • Bull NovaScale R423-E2
    • 360 TB user data (2 x DDN S2A9900 storage)
  • Aggregated data rate ~20 GB/s

Infrastructure Nodes

Login Nodes

  • 12 x Bull NovaScale R423-E2 (8 cores, 24 GB DDR3)

GPFS Gateway Nodes

  • 6 x Bull NovaScale R423-E2 (8 cores, 24 GB DDR3)

Management Nodes

  • 2 Master Nodes with Master Repository
  • 35 Admin Nodes for scaled system services

Software

  • SUSE SLES 11 Operating System
  • ParaStation Cluster Management
    • GridMonitor
  • Torque/Moab Batch and Resource Management System
  • Intel Professional Fortran, C/C++ Compiler
    • Intel Cluster Tools
  • Intel Math Kernel Library
  • ParTec MPI Message Passing Interface
  • OpenMP Intra-Node Programming Model

last change 07.07.2009 | Ulrich Detert | Print