Information Technology and Communication (ITC) purchased over 200 nodes for a Linux cluster (dogwood.itc) in 2006 . Researchers considering their own purchase of a Linux cluster are invited to consider the purchase of nodes on this ITC Linux cluster instead. This is the second year of a program that enables researchers to purchase nodes in a Linux cluster that ITC houses and maintains for a three-year period. Some of the benefits include:

• Free system administration
• Timely security updates
• 24 hour a day monitoring in ITC's machine room
• Ability to use more nodes than purchased during periods of non-peak demand.

Program Details

Cluster and node details

• Minimum order is four nodes - you are encourage to buy at least eight nodes.
  The per node cost is $XXX.
• The nodes are aDataZapper 2Ks with two YY-Ghz XAMDeon processors with Z GB of L2 cache per CPU and 4 GB of RAM per node. The purchase includes a three-year warranty on each node
• Interconnect among nodes will be Infiniband and cluster dedicated/restricted to parallel computing.
• ?? --> Initial Operating System (OS) will be for 32-bit operation and applications.
• Purchase includes a three year warranty on each node.
• Will be operated as a production cluster, installed in the ITC data center in Carruthers Hall, which includes uninterruptable power supply (UPS) and backup generator power and a 24-hour per day operation.
• ITC uses NPACI Rocks cluster administration suite, which includes the PBSPro batch queuing and job submission software. This is what is used currently on ITC's other three Linux Clusters, Birch, Cedar, and Dogwood.
• One-time cost for the initial purchase of your nodes covering the three-year life of the cluster - no annual or recurring charges.

Faculty may purchase nodes under one of two plans:

Priority Access:

• Individual nodes will be part of the whole cluster. Guarantee of priority access to at least 80% of the cycles of your nodes. Note: Priority access means that your jobs will go to the front of the job queue when submitted. PBSPro scheduler will be used in a non-preemptive mode which means that if all the nodes are in use when your job is submitted, you will have to wait for jobs to complete before your's starts.
• Maximum job submits at one time, maximum job CPU time, etc is determined by ITC for the benefit of all users.

Exclusive Access:

• Individual nodes are separate queue from the whole cluster. No other users than those you allow can run jobs on these nodes.
• User queue priority, maximum jobs submits at one time, maximum job CPU time, etc. will be determined by faculty purchaser.
• Access through the PBSPro scheduler to all the nodes in the general queue, without any special priority.

ITC Provides for either purchase scheme:

• Electrical power, (including uninterruptable power supply), machine room space, and cooling.
• Network capacity
• System software, clustering software, system administration, setup, configuration, and the usual research applications (e.g,. Matlab, Fortran 90).
• A pool of local spare parts to ensure that you always have access to the nodes you fund.
• Local, temporary storage for use during computational runs. (Note: You are NOT purchasing any long-term storage space as part of your nodes.)
• A minimal amount of system administration time to install a user-purchased application on the cluster for software packages ITC deems to be of general use and licensed appropriately. Complex or problem installations will be done on a for-fee basis. Researchers are strongly encouraged to talk to Research Computing Support staff (email res-consult@virginia.edu or call 243-8800) before making a software purchase.
• The procurement (purchase) of the cluster and nodes, including compliance with SWAM and eVA guidelines as well as receipt of cluster, unpacking, and setup. Nodes purchased with ETF funds will have included appropriate as well as free storage after the three-year life cycle. ITC can arrange the inventory management to ensure departments will not lose their ETF allocations by participating in this project..

Important Details

• The per-node cost includes the cost of the computer node, its three year maintenance contract, and pro-rated partial cost of the network switch and chassis, KVM and equipment rack. It does NOT include any charge for the cluster's head node, local temporary storage, the HVAC, electricity use (uninterruptible power supply), machine room space, and system administration -- ITC is providing all these as its contribution. (Note: node cost is only the incremental cost of adding nodes to this cluster. It is not the fully loaded cost share of the full cluster purchase and operation.)
• You are free to pull out at any time during the three year cycle. You will receive your compute nodes only. We cannot return the pro-rated share of the cluster's common equipment.
• At the end of the three year life cycle of the cluster, you may opt to take your nodes or if you prefer, we will surplus them for you. If your nodes were purchased with ETF funds, we will store those nodes for free.
• Longer-term storage will need to be provided by the researcher (departmental storage, lease of disk wedges from ITC, etc.). ITC can help you determine the best way to provide additional storage if you need more disk storage as part of this purchase.
• ITC will have the cluster ready for production runs no later than summer 2006, assuming there are no shipping delays after the order is placed in March 2006.
• When you place your order for cluster nodes, you agree to the these terms and conditions.
• Deadline for request to participate in this purchase is Friday, March 24, 2006.

Researchers are encouraged to contact Tim Tolson (tft8g@virginia.edu), the manager of ITC's Research Computing Support Group to discuss their individual needs and obtain a customized cost analysis.

Problems submitting the form- email Tim Tolson or phone him at 243-6592.
[an error occurred while processing this directive]