EDDY is one of the largest HPC clusters in Europe dedicated exclusively to wind energy research and is operated by the IT Services of the University of Oldenburg. It is funded by the Federal Ministry for the Environment, Nature Conservation and Nuclear Safety (BMU) through a programme aimed at increasing the share of renewables in Germany’s total electricity generation. The users of the cluster are

• the Complex Fluid and System Dynamics project group at the Fraunhofer IWES Institute.
• Energy meteorology (EnMet)
• Turbulence, Wind Energy and Stochastics (TWIST)
• Wind Energy Systems (WeSys)

The cluster is used for challenging computational fluid dynamics (CFD) calculations across a wide range of (spatiotemporal) scales, from individual blades and wind turbines, to the simulation of wind farms, to mesoscale and weather models.

Hardware Overview

• 244 compute nodes (5,856 CPU cores, 21 TB of main memory (RAM), 201 TFlop/s theoretical peak)
  • 160 ‘low-memory’ compute nodes
    • LENOVO NeXtScale nx360 M5 HPC Node (Model 5465-FT1)
    • 2x Intel Xeon CPU E5-2650 v4 12-core at 2.2 GHz
    • 64 GB main memory (RAM) @ 2400 MHz
  • 81 ‘high-memory’ compute nodes
    • LENOVO NeXtScale nx360 M5 HPC Node (Model 5465-FT1)
    • 2x Intel Xeon E5-2650 v4 12-core CPUs at 2.2GHz
    • 128 GB main memory (RAM) @ 2400 MHz
  • 3 ‘GPU’ compute nodes
    • LENOVO NeXtScale nx360 M5 HPC Node (Model 5465-FT1)
    • 2x Intel Xeon E5-2650 v4 12-core CPUs at 2.2GHz
    • 56 GB main memory (RAM) @ 2400 MHz
    • 1 TB 7.2k 6 Gbps HDD (used for local storage)
    • NVIDIA GPU

• Management and Login Nodes

  • 2 administration nodes in an active/passive high-availability (HA) configuration

    • LENOVO System x3550 M5 Node Model 8869-FT1
    • 2x Intel Xeon CPU E5-2650 v4 12-core at 2.2GHz
    • 256 GB main memory (RAM) @ 2400 MHz
    • 4x 1.2TB 7.2K 12Gbps SAS HDD (configured as a hardware RAID 10)
    • Connect-IB single-port card

    
    The master nodes are shared between CARL and its sister cluster EDDY and run all vital cluster services (node provisioning, DHCP, DNS, LDAP, NFS, Job Management System, etc.). They also provide monitoring functions for both clusters (with automated alerting). Monitoring covers hardware components (health status of all servers, temperature, power consumption, etc.) as well as basic cluster services (with automated restart should a service fail)
    

  • 2 login nodes for user access to the system, software development (programming environment), and job submission and control. These nodes have the same specifications as the administration nodes described above.
  • Internal networks
    • An InfiniBand network comprising 9 spine switches and 15 leaf switches with a 1:1 (non-)blocking factor. Consequently, each leaf switch is connected to 18 MPC nodes. The maximum data transfer rate is 56.250 Gb/s (4x FDR).
    • Secondly, a physically separate Gigabit Ethernet network (‘base network’) for vital cluster services (node provisioning, DHCP, DNS, LDAP, NFS, Job Management System, etc.)
    • A 10 Gb Ethernet backbone network connecting the management and login nodes, the storage system, and the Gigabit Ethernet (MPI and base network) leaf switches
    • IPMI network for hardware monitoring and control, including access to the VGA console (KVM functionality), enabling full remote management of the cluster
  • Storage System
    • General Parallel File System (GPFS) with 1.392 PB of total storage space, of which approximately 926 TB is usable. We are using four declustered arrays to ensure the availability of our data. The failure of a single hard drive goes unnoticed; even if two hard drives fail, the ‘critical rebuild’ would take only about 45 minutes. The high-memory and pre- and post-processing nodes have additional local storage space (up to 1 TB).
    • Enterprise-class scalable NAS cluster (manufacturer: EMC Isilon). This is where the home directories are stored. All data saved on the Isilon is backed up, and it is possible to work with snapshots. The data is accessible via a 10GbE connection.
      
      The Isilon serves as the central storage for IT Services, which is why it is also used for HPC. Disk space is allocated to the two clusters depending on the proportion of the storage system’s hardware that was funded by the FLOW and HERO project funds, respectively.

SYSTEM SOFTWARE AND MIDDLEWARE

• All cluster nodes are running Red Hat Enterprise Linux (RHEL)
• Cluster management: Bright Cluster Manager
• Commercial compilers, debuggers and profilers, and performance libraries: Intel Cluster Studio, PGI
• Workload management: SLURM

SELECTED APPLICATIONS RUNNING ON EDDY

(Due to licensing restrictions, some applications are only accessible to specific users or research groups.)

• PALM (an advanced and modern meteorological model system for atmospheric and oceanic boundary-layer flows)
• OpenFoam (a toolbox for the development of customised numerical solvers for computational fluid dynamics)
• MATLAB, including the Parallel Computing Toolbox
• WRF (a mesoscale numerical weather prediction system designed for atmospheric research and operational forecasting applications)

PICTURES

More pictures can be found here.