Linux Clustering - Hardware Configuration

Pratt Central Computing Cluster

Our goal is to provide Pratt with an advanced cluster infrastructure to meet the growing needs of our researchers. While Duke is moving forward with a central cluster infrastructure, we hope to provide a system where researchers and IT staff can experiment with advanced clustering concepts as well as running production jobs locally. The goal is to complement any existing central infrastructure, not duplicate it.

Our initial goal to to obtain a 4-6 node SMP Athlon K7 cluster as well as a 4-node Opteron cluster.

Switch Fabric

We need to ensure we have a switch fabric that can handle future cluster communication loads. Potential solutons include something like an HP ?ProCurve 4100 or 5300 switch. Use GigE for the master nodes and Full-Duplex 100Mbps ports for the nodes. Whatever switch we use must have SSH capability for potential remote mgmt and configuration.

Master Node(s)

Dual Athlon or Pent 4 processors (Opteron eventually)
Motherboard with advanced chipset and large memory capacity (4-8GB)
64-bit/66MHz PCI-X for GigE, Fibre Channel NICs
2U form factor with 3 PCI slot riser
Will RAID-0 make any difference? Where would the disks be a bottle neck?

Cluster Nodes

Dual Athlon or Pent 4 processors (Opteron eventually)
Motherboard with advanced chipset and large memory capacity (4-8GB)
1GB Memory
1U Form factor if possible
On board GigE NIC
Dual HD capabiltiy (for potential RAID-0 configuration)
1 PCI riser slot (PCI-X would be nice, but is not necessary)