Neuronic

neuronic.nic.uoregon.edu

Note: Neuronic is long gone. This page is kept for reference only.

Basics

• 2U Dell PowerEdge 2650
• 34 i686 Xeon Processors:
  • Frontend: 2 Single-Core Xeon (Prestonia) @2.8Ghz
  • Nodes 0-15: 2 Single-Core Xeon (Prestonia) @2.8Ghz
• 68 Gb of RAM:
  • Frontend: 4Gb
  • Nodes 0-15: 4Gb
• SCSI Disks:
  • Frontend: PERC 3Di RAID; 36Gb mirror + 73Gb mirror
  • Nodes 0-15: 36Gb
• Myricom Myrinet-2000 Cluster Interconnect (M3S-PCI64C 2MB)
• 2 onboard Broadcom Tigon3 copper NICs
• Dell Embedded Remote Access Controller III for remote management
• Rocks Cluster Linux 5.0 (CentOS 5.2 + perfctr kernel)
• All nodes in one full-height Dell rack

Detailed Configuration Info

Consult the Configuration subpage.

Load Monitoring

You may check the cluster utilization and job queue with Ganglia on Mist.

MPI and Torque+Maui

Getting Started

See Software:PBS for general information on using our Torque+Maui batch scheduling system for interactive or parallel jobs. Included are instructions on how to compile your code, write a batch job, and submit it for processing.

Queues

Jobs submitted from the Mist frontend will enter into the 'neuronic' queue by default. This contains all 16 nodes. See Queues for a list of available queues.

Myrinet High-Speed Cluster Interconnect

All Neuronic compute nodes are equipped with Myrinet cards that give double the throughput and less than 1/4 the latency of Gigabit Ethernet. Any MPI-enabled programs built using the scheduler-enabled MPIs will automatically use these interconnects.

Node Attributes

The Neuronic nodes are all identically configured. Unlike Mist, there are no special attributes available on a subset of the nodes. See Batch Attributes for more information about using qsub arguments or job script lines to fine-tune your run parameters.