In general, the cluster configuration should remain as “standard” as possible; this is not be a bleeding edge venture.

Some general information gleaned from our anticipated user base:

• Software: mostly home grown code, C/C++ and Fortran programs (probably a Portland suite of compilers), Matlab (and MatlabMPI), AMBER, NAMD, CHARMM, and Gaussian.

• 64 bit compilations will be required, as is a 64-bit operating system with 32-bit backward compatibility

• Most application programs will be floating point and CPU intensive versus relatively few integer and data intensive applications

• A Linux operating system is required.

• A ROCKS based implementation is preferrable with PBS as the scheduler (at least most users had experience with this).

• MPI will be required, preferably to as many work nodes as possible

• Any other cluster management tools for monitoring, administration, etc., will be a plus.

Here is a rough hardware configuration guide …

• 1 head node (Dual Intel 5148 processors, 2 GB RAM, mirrored 80 GB HD, 2 channel FC HBA) on private & public network

• 1 NFS node (Dual Intel 5148 processors, 2 GB RAM, mirrored 80 GB HD, 2 channel FC HBA) on private & public network

• 60 lightweight work nodes (Dual Intel 5148 processors, 2 GB RAM, single 80 GB HD (or smaller)) on private network

• 4 heavyweight work nodes (Dual Intel 5160 processors, 16 GB RAM, single 80 GB HD (or smaller)) on private network

• Sixteen of the lightweight work nodes should be interconnected via Infiniband for low latency. The rest should be interconnected via gigabit ethernet.

• Although local disk space is available, our NAS/SAN filesystems will provide the storage area for datasets and home directories

The quote should include all racks, power distribution, and UPS equipment needed to house and run the cluster. The quote should also include the network switch equipment required for internode communication.

We will want the equipment to be assembled, racked, and imaged by the vendor either onsite or prior to delivery (no preference). We will require 24/7/4hr support for the master and NFS nodes, and 24/7 support for OS/software. NBD support will be sufficient for the compute nodes. The duration of all support should be three years. After delivery/installation, we will require cluster administration training/knowledge transfer for our UNIX administrators. All these costs should be reflected in the quote.

Home