The CM department runs a computer cluster with the following parameters:

• RAM: 5 TB Ram HDD: 100 TB CPUs: 2200 Opteron Cores:
• (a) 235 x Intel Xeon Quad Core, 3.0 GHz, 2 CPU per node
• (b) 80 x Opteron 280 DualCore, 2.4 GHz, 2 CPU per node
• (c) 160 x Opteron 250, 2.4 GHz, 2 CPUs per node
• Network: 20 GB Infiniband (MPI), GB Ethernet (Service)
• Queuing system: Sun Grid Engine
• OS: SuSE Linux 10.3
• Administration tool: scVenus

In our cluster a state-of-the-art water-cooling technique has been applied. In contrast to conventional air-conditioned computer centers, where the air of the entire room has to be cooled, here only the air inside of a computer rack is conditioned. That results in a higher cooling efficency. The installed cooling equipment controls both temperature and humidity inside the racks.

By taking advantage of the slash-5 technique our nodes are packed with a doubled density within the racks.

As working PCs we mainly use hardware with

• RAM: 1-4 GB
• CPUs: 1-2 x Opteron, 2.4 GHz, dual and quad core
• Graphic: nVidia OpenGL
• Network: GB Ethernet
• OS: Linux (Windows remote) or Windows (Linux remote)
• Administration tool: scVenus

Users are free to choose between Linux and Windows. Linux users can remotely control Windows sessions via rdesktop whereas Windows users can connect to Linux via X-Win32.

Our PCs are managed remotely using the cross-platform administration tool scVenus. New PCs can be unattendenly installed. At boot time only the local operation system (Linux or Windows) has to be chosen manually. After approx. 1 hour the PC has downloaded and installed our default software selection, the user accounts have been created, the PC has been incorporated into our network, and is ready to use.

All home directories of all compute nodes as well as the PCs are cross-mounted with a network appliance system:

• storage capacity: 10 TB
• reliability: high availability (HA) solution
• RAID Level: 2 x 6
• snapshots: 1h, 1d, 1w, 1m
• protocols: NFS, CIFS

The installed network appliance system is equipped with fast caches to create snapshots on the fly without any loss of performance. Every directory in the user's home/ folder contains the read-only and hidden snapshot folders ".1_hour_ago", ".1_day_ago", ".1_week_ago", ".1_month_ago". Each folder contains the files from the corresponding time/dates. Unintentionally deleted files can be easily restored by copying them from one of the snapshot folders.

We run two master nodes in a HA setup (1 node running actively, 1 node running in stand-by). Each master has the following setup

• CPUs: 2 x Opteron, 2.8 GHz
• RAM: 16 GB
• availability: 2 redundant power supplies
• controller: HP SmartRaid

The combination of both NAS and RAID systems offer different reliability levels at different costs.