|
Hardware Description
In terms of hardware, the EHPCC is based upon 64 Sun v20z 1U computers each of which contains two AMD Opteron 2.2 Ghz processors which enables a 64-bit computing environment with 32-bit compatibility. The AMD Opteron processor assists application performance by reducing memory latency using "Direct Connect Architecture". Each compute node is equipped with an Ultra320 SCSI tranpsort which results in faster disk access times and longer times between disk failures. Twenty one of the nodes have a Myrinet card to provide high data rate and low latency communication especially for particular applications that might be network bound. Here is a summary of the cluster's key characteristics:
| Vendor |
Sun Microsystems |
| CPU |
Opteron 2.2 Ghz (2 CPUs per node) |
| Memory |
2 GB DDR1/333 (4 GB in Parent node) |
| Transport |
SCSI Ultra320 |
| Storage |
72GB per worker node, 300GB on Parent node |
| Network |
Dual Gig-E NICs on each node; 21 nodes have a Myrinet interface |
| OS |
RedHat Enterprise Linux WS 3.0 AMD64 |
| Job Mgr |
Sun N1 Grid Engine 6 |
| Master Node |
1 V40z, 4GB RAM, 300GB storage |
| Worker Node |
63 V20z, 2GB RAM, 72GB storage |
The 64 nodes are distributed across two racks, each of which contain 32 v20z, 1U computers. Both racks contain terminal management devices, thermal probes to insure operation within specified thresholds, GigE and Myrinet network switches, network cabling, power distribution units, and heat removal appliances. This installation is housed in the North Decatur Data Center located on the third floor of the North Decatur Building.
Networking within the cluster racks is accomplished using a combination of SMC switches to support the internal job distribution network as well a "service processor" network to collect information on environmental operations (e.g. temperature, fan speed,etc). There is also a Myrinet switch and associated technology to provide a low latency network, |
 |
|
