Lakemead Itanium Cluster

Description

This High Performance Itanium Cluster is a research resource, intended for testing and running large codes, distributed parallel-processing codes, large memory codes, and it is also a test-bed for research groups who may want to explore some features and applicability of cluster computing to their respective research areas.

 
lakemead
Machine Type Itanium-2 Cluster
Operating System GNU/Linux
Kernel 2.4.21-37.EL
Processors 64 Intel IA-64 1.3GHz
Total Cluster Nodes 32 dual nodes
1 Master Node
1 Development Node
1 Node (internal use)
29 Compute Nodes
Network Gigabit Ethernet, Fast Ethernet
Memory 256GBytes (32@8GB)
Storage Approx. 14.4 TBytes

The Hardware

  • This Intel Itanium-2 cluster consists of 32 dual IA64 nodes.
  • Each node has 2 Intel IA64 processors with a clock speed of 1.3GHz.
  • Each processor can access its own memory and that of the other processor in its node. Therefore, this is a 2-way SMP processor (8 GBytes of shared-memory).

Node Interconnect

  • All nodes are connected using Gigabit Ethernet, 10 times faster than 100 Mbit ethernet. Read a "white paper (pdf)" on Gigabit ethernet at the intel site

Memory

  • There are 8 GBytes of shared-memory per node
  • Total amount of memory (distributed) is 256 Gbytes
  • For more details about the Itanium 2 processors take a look at this Intel site

File Systems

  • Each compute node has 450 GBytes of disk space, for a total disk capacity of 14.4 TBytes.
  • $HOME directories are NFS-mounted to all nodes. There is only 140GBytes of space on this file system.
  • Each node has fast local disk space called /scratch. This file system has a capacity of about 270 GBytes. Users are encouraged to perform most of their I/O using this file systems, in particular for parallel jobs.
  • For parallel jobs I/O can be done to the home directories but be aware that performance is going to be affected.
  • There are examples on how to copy files before and after your job starts running.

Configuration

  • lakemead has a master or head node to do most interactive work, like editing files, submitting jobs, checking job status.
  • There is a node called development whose primary function is to be used for code development, code compilation, debugging.
  • All compute (29) nodes must be accessed in batch mode using SUN Microsystems Sun-Grid-Engine (SGE)
  • One node is reserved for internal projects

User Environment on lakemead

  • Permanent /home space
    Your $HOME directory is allocated ??? GB. HOME directories are backed up..
    Temporary space
    users are encouraged to use the fast local disks on each node as temporary space (/scratch)