Cluster Computing

IMI HPC cluster was established to cover the institutes need to calculate heavy computational tasks. The primary use is to develop and calculate computational algorithms for picture analysis, segmentation and nucleotyping.

Overview

Our HPC cluster is based on open source Rocks Cluster v.5.1 witch is based on CentOS 5.2 64 bit Linux OS. The solution has 1 Master-node and 24 client-nodes, the Master-node has two network cards. One of them is connected to the nodes (node network), and the other to the research network. The WEB portal in Rocks allows the user to administrate and supervise the resources and jobs running in the cluster.

Screenshot from Rocks web portal

Node handling

Data is scanned and produced on the research network and saved on the file server. The file server, which is also available on the node network, allows the cluster nodes to read data directly from the windows share, giving a high data throughput.

The Master-node can reinstall the Client-nodes when needed. This installation and reinstallation of Client-nodes is controlled by Kickstart, developed by Red Hat.

To be able to control the jobs and supervise the task at hand the cluster has the following components installed

• Distributed resource manager: SGE Sun Grid Engine
• Surveillance Web interface: Ganglia

The Master-node runs on a HP BL460c blade server with Intel Quad Core Intel Xeon E5450 (12MB Cache, 3,0GHz, 1333MHz FSB) 8GB RAM and 6×146GB 10k SAS HDD.

The Client-nodes were built at IMI’s Technology unit and consist of 12 Supermicro Twin 1-U rack unit with 2 Quad Core Intel Xeon E5420 (12MB Cache, 2,5 GHz, 1333MHz FSB), 2×8GB RAM og 320GB 7.2k SATA HDD.

Total available CPU kernels: 192