Support for non-DSC HPC systems

The DSC plays a major rule in the support of FSU's computational infrastructure by providing expert technical support in advanced computing. In addition to the management of FSU's shared multidisciplinary HPC (see http://hpc.fsu.edu), the DSC has also invested a large sum of funds to develop a computing facility located in the center of the main FSU campus. This facility is currently equipped with 1000 square feet of raised floor, battery backup power, high-speed network connectivity to the campus network backbone and the Florida Lambda Rail, and 80 tons of cooling.

As part of it's service to University, the DSC hosts computational servers owned by other FSU academic units on a first-come, first-served basis until 25% of the DSC machine room cooling, electrical, or floor capacity is reached. Once the capacity of the room is reached, the DSC will not support additional machines.  

Eligibility

All units within the College of Arts and Sciences are eligible to have machines hosted in the DSC computer room.  The DSC's Local Systems Committee will evaluate each request on a case-by-case basis and will agree to host machines provided the capacity allocated for non-DSC systems has not been reached and the requirements listed below are met.

Requirements

Each of the requirements enumerated below is described in greater detail in the following pages.

1. The DSC is not responsible for incidental costs associated with the installation or maintenance of computer systems hosted in the DSC computer room.
2. Requests to house computer systems in the DSC computer room must include a detailed hardware proposal.
3. Requests to house computer systems in the DSC computer room must include a proposal describing how the computer systems will be managed.
4. Computer systems hosted in DSC computer room are required to run the DSC load distribution system.
5. Requests to house computer systems in the DSC computer room must include a designated faculty liaison from within the unit making the request.

For questions or information regarding availability contact Jim Wilgenbusch at 645-0307 or This email address is being protected from spambots. You need JavaScript enabled to view it..

1. Incidental costs
In addition to the substantial costs of providing a cold room with reliable power, there are a number of "incidental" costs associated with hosting computer systems.   These costs include server racks, power distribution units (PDUs or power strips), and network switches.  The DSC is not responsible for covering incidental costs.  At a minimum, the DSC will provide a single outlet for each electrical circuit required to power the computer system and a single 10/100/1000 network uplink in the vicinity of the computer system.  The cost of incidentals (e.g., racks, PDUs, and network connectivity) can either be included in the price of a fully integrated system or the DSC will provide these items at cost.  In order to avoid half empty racks, the later option may be required. For example, if a group plans to purchase a small number of servers, then the DSC would prefer to use an existing rack, PDU, and network switch.  In this case, costs will be assessed according to the portion of the resource being used.   

2. Hardware proposal
Requests to house computer systems in the DSC computer room must include a detailed hardware proposal.  An agreement to house a computer system in the DSC computer room is limited to the hardware described in the proposal.  This requirement is intended to safeguard against someone purchasing hardware that is in some way incompatible with the DSC machine room infrastructure.  In addition, the hardware proposal is needed to estimate the power, space, and cooling requirements of the computer system so that we do not exceed the capacity of the DSC computer room.  The DSC has a wealth of experience with numerous hardware vendors and can create the hardware proposal on the behalf of unit making the request.

Hardware proposals must include the following information:

• Server rack dimensions *
• Switch type and size *
• PDU type and quantity *
• Chassis type per unit
• CPU type and quantity per unit
• Motherboard type per unit
• Ethernet or network adapter type and quantity per unit
• Hard drive type and quantity per unit
• Memory type and quantity per unit
• Type and quantity of all peripheral devises that will be part of the computer system (e.g. disk arrays, CD/DVD drives, etc.)

* if not using equipment provided by the DSC

3. Management proposal
Requests to house computer systems in the DSC computer room must include a proposal describing how the computer system will be managed.  Two general administrative options are available; 1. the DSC will manage the computer system, and  2. the unit making the request will manage the computer system.

Option 1: DSC manages the computer system
Option 1 gives a group access to a computational resource without the hassle of having to administer it.  The DSC will agree to manage a computer system only if the DSC is given discretion to choose the hardware, the operating system, and the tools used to update and distribute software on the computer system. The DSC systems manager determines who is granted access to privileged commands. Access to privileged commands will be controlled by the "sudo" command; root logins are not permitted. Only members of the DSC systems group will know the root password. The DSC systems manager will determine the update cycle of the core operating system and system software.

Option 2: Non-DSC unit manages the computer system
If a unit chooses to manage their own computer system, then a qualified systems administrator must be designated to manage the resource.  The required management proposal must address in detail how the designated systems administrator will tend the following tasks: account management, disk management, network management, system installation, software management, security, systems management, user support, hardware support, and administration of purchase and warranty information. Systems not managed by the DSC will be relegated to a separate secure network.  A firewall will monitor incoming and outgoing network traffic and the DSC systems manager will determine the stringency of the firewall. Access the machine room will be limited to the designated systems administrator.  Members of the DSC systems group will have root access to all systems housed in the DSC computer room.

4. Cycle sharing
In order to maximize throughput on DSC and non-DSC computational servers and cluster nodes, the DSC uses a job scheduling and queuing system called Condor (http://www.cs.wisc.edu/condor/). All of the computer systems housed in the DSC computer room must participate in this system.  The Condor system allows fine-grain configuration of policies to support distributively owned computing resources.  For example, specific nodes can be configured to "prefer" jobs submitted by specific users. This way, a resource owner will have limited access to all of the DSC computational resources and guaranteed access to his or her machines. This system ensures that cycles are not wasted if the primary resource owner is not using them.  

While the Condor system provides some important advantages to managing and distributing computationally intensive jobs, a resource owner is not obligated to use it.  For example, a resource owner can run jobs interactively on any of his or her computer systems and his or her jobs will preempt jobs managed by the Condor system.  Circumventing the Condor system in this way is to the disadvantage of the resource owner, however.   By submitting jobs through the Condor system, a resource owner will not only have guaranteed access to his or her machines, but the Condor system will also provide access to the other machines housed in the DSC machine room.

5. Faculty Liaison
Requests to house computer systems in the DSC computer room must include a designated faculty liaison from within the unit making the request. The liaison will be responsible for conveying information to the users of the computer system housed in the DSC computer room.  If the DSC agrees to manage a computer system, the faculty liaison is responsible for approving account requests, requesting changes to a user's priority, and requesting software installations.  The authority to make these requests can be transferred to a designate person (e.g., graduate student) at the written request of the faculty liaison.