NOTE: Cluster in normal productional use
After the extension, the Alibaba cluster now is
a Linux-cluster from Dell with 480 cores in 80 workernodes (Intel Xeon CPUs), a parallel lustre-filesystem and
a performant infiniband network.
The nodes are called:
- mardschana: grid-software
- sesam: fileserver (no direct access for the user)
- abdallah: admin-server (only used by admins)
- r00 - r39 (2 dual-core Woodcrest processors per node, 2 GB of main memory per core, gigabit-ethernet interconnect)
- r40 - r79 (2 quad-core Harpertown processors per node, 2 GB of main memory per core, infiniband interconnect)
The worker nodes can be accessed only by using the batch system (Torque).
The functional view of the cluster is presented in the following picture :
The cluster is situated in the computing center of the ZIB, which is a highly secured area.
All critical services are driven by uninteruptable power supply.
The hardware is distributed to four racks, according to the following plan:
Before the extension, the Alibaba cluster was ...
a Linux-cluster from Dell with Intel Xeon CPUs, 182 cores with 400 GByte main memory and doubled gigabit-ethernet-access (1.940 GFlop/s peak):
The cluster consisted of five server nodes, 40 worker nodes and an additional admin node.
The operating system was Suse Enterprise Linux v10 EM64T
on the server nodes
and OpenSuse 10.1.
on the worker nodes.
- 10 Mar 2008