Former national supercomputer HP XC4000
Between 2007 and 2013, the SCC operated an HP XC4000 system as the state supercomputer for Baden-Württemberg. The massively parallel computer (MPP system), which is based on AMD Opteron processors, was available for projects from the state's universities that required computing power that could no longer be met by local resources, but could also be used by business and industry.
Configuration of the HP XC4000 system
Nodes on the HP XC system could take on different roles. In Karlsruhe there were
- 2 login nodes with 8 cores each, with each node providing a theoretical peak performance of 41.6 GFLOPS,
- 750 compute nodes with 4 cores each , with each node providing a theoretical peak performance of 20.8 GFLOPS,
- 2 computing nodes with 8 cores each, with each node providing a theoretical peak performance of 41.6 GFLOPS,
- 8 service nodes for resource management etc., each with 4 cores,
- 10 file server nodes with a net disk capacity of 56 TB and
an InfiniBand 4X DDR interconnect as the connection network.
The HP XC4000 system was a massively parallel "distributed memory" parallel computer with a total of 772 nodes, all nodes (except for the file server nodes) had a clock frequency of 2.6 GHz. All nodes had local memory, local disks and network adapters. A single computing node had a theoretical peak performance of 20.8 GFLOPS or 41.6 GFLOPS, resulting in a theoretical peak performance of 15.77 TFLOPS for the entire system. The main memory across all computing nodes amounted to just over 12 TB. All nodes were interconnected via an InfiniBand 4X -->interconnect. -->
The base operating system on each node was HP XC Linux for "High Performance Computing"; this was a Linux implementation compatible with RedHat AS 5.1. Above this was the XC parallel software environment, which allowed the user to use the parallel computer efficiently.
The scalable, parallel file system Lustre was used as the global file system within the XC system. By using several Lustre Object Storage Target (OST) servers and Meta Data Servers (MDS), both high scalability and redundancy in the event of individual server failures were achieved. 56 TB were made available for Lustre. In addition, each node of the cluster was equipped with local disks for temporary data.
Detailed brief description of the nodes:
- 2 eight-way (login) nodes, each with 4 AMD Opteron dual cores (HP ProLiant DL585) with a clock frequency of 2.6 GHz, 32 GB main memory (DDR-400) and 288 GB local disk space,
- 750 four-way (computing) nodes, each with 2 AMD Opteron Dual Cores (HP ProLiant DL145 G2) with a clock frequency of 2.6 GHz, 16 GB main memory (DDR-400) and 144 GB local disk space,
- 2 eight-way (computing) nodes, each with 4 AMD Opteron Dual Cores (HP ProLiant DL585) with a clock frequency of 2.6 GHz, 128 GB main memory (DDR-400) and 8*144 GB local disk space,
- 8 four-way service nodes, each with 2 AMD Opteron Dual Cores (6 HP ProLiant DL385 and 2 HP ProLiant DL585 nodes) with a clock frequency of 2.6 GHz, 8 GB main memory (DDR-400) and 144 GB local disk space,
- 10 two-way (file server) nodes, each with 2 Xeon cores (HP ProLiant DL380 G4) with 36 connected SFS20 enclosures, which provided 56 TB net disk space.