| Introduction
Electronic and Computer Services (ECS) maintains a parallel-computing cluster for use by faculty and students of the College of Engineering. The primary purpose of the cluster is to support the College’s academic mission. Undergraduate and graduate course support throughout semesters will be the primary use of the system. However, the cluster is available for research computing needs when the load generated by courses is low.
System Configuration:
- Head Node – Dual 3.4GHz Dual-Core Processor Configuration w/ 8GB DDR2 RAM and 1.2TB 10K RPM Hard Disk Storage
- Nodes – 24 Dual 3.4GHz Processor Configuration w/ 4GB DDR2 RAM each, and 40GB 7.2K RPM Hard Disk Storage
- Connectivity – 100 Mbps Backbone Connectivity and Cluster Nodes Interconnections Operating at 1 Gbps
- Operating System – Linux
Appropriate Use
Priority use of the cluster will be given to students enrolled in Engineering courses that faculty have identified and registered as requiring high performance computing; the course identification/registrations process will occur each semester. Students utilizing the system for course work may only use it for its intended purpose, work directly related to the class being taken. Reading e-mail, web browsing, document preparation and other “normal” computing tasks should be performed on other systems more appropriate to those tasks.
The system will be available to research programs on an as-available basis during semesters. Faculty will need to provide the names of students so that they can be added to user access list; unless indicated otherwise by a faculty member, this access will cease at the end of a given semester. If long-term access is required, the faculty member will need to specify the length of time access is required.
All use of the College of Engineering cluster will adhere to current College and University computing policies.
Account Procedures
Faculty in the College of Engineering will automatically be given account access to the system; account logon will be through an individual’s Access Account and Password. Faculty desiring to use the College High Performance Computing system will need register for system access by visiting http://www.engr.psu.edu/HPC_Registration. Once registered, a faculty member will be allocated 1Gbyte of disk space for data storage. These accounts will be audited annually in June to determine if access to the system is still desired. Disk space of faculty not desiring to use the system in the following fiscal year or failing to respond to the audit will relinquish allocated disk space; relinquished space will be erased and reallocated annually No Later Than (NLT) 15 July.
Faculty desiring to use the College High Performance Computing system for scheduled classes will need register for system access by visiting http://www.engr.psu.edu/HPC_Registration. By providing the requested course information, faculty will ensure that all students registered for the class are granted access to the system. Course registration information will be automatically reviewed and updated daily until the University’s Drop/Add period closes; this will ensure that only those students officially registered will have access to the system. Each student registered will be given 500Mbytes of temporary disk storage on the system. The temporary disk storage will be available to student throughout the semester and no more than thirty (30) days after the last day of final exams, at which time the disk space will be erased and made available for the next semester’s use.
Faculty desiring to afford graduate or Honors Thesis students with an opportunity to use the College High Performance Computing system will need register for system access by visiting http://www.engr.psu.edu/HPC_Registration. Faculty will be expected to estimate the length of time a student will require access to the system. Once registered, students in this category will be allocated 1Gbyte of disk space for data storage. These accounts will be audited annually in June to determine if access to the system is still desired. Faculty will need to annually revalidate a student’s access requirements to the system. Any student use requirements not revalidates will relinquish allocated disk space; relinquished space will be erased and reallocated annually NLT 15 July.
Software Installation
The College’s faculty Academic Computing Committee (ACC) membership that includes representatives from each department will function as the governing body of this high performance computing resource. ACC faculty members will determine/approve the use of specific applications on the cluster.
The College will continue to provide base application support by purchasing the annual fifty-seat licenses cost for the MatLab, Simulink, Symbolic Math Tools, Distributed Computing Toolbox, and the Distributed Computing Engine. Purchase of department/discipline specific licenses for toolboxes will be the responsibility of individual departments or cost sharing among departments.
Commercial vendors offering deeply discounted software licenses typically place strict licensing use limitations on applications that are purchased for academic use. In these cases, the licensing agreements specifically prohibit use of their applications for other than class related work. These restrictions typically apply to software that may be used for commercial and research purposes; e.g., Mathworks’ licensing agreement is once such application. When an application installed on the cluster has its use constrained by a license agreement, researchers/departments may procure appropriate license that will allow for the restricted licensed software use for other than classroom instruction. Conversely, licensing use restrictions do not typically apply when research license are used for educational purposes.
Costs associated with the procurement of other ACC approved academic/research applications will be the responsibility of individual departments or faculty research programs; this also applies to recurring licensing costs. Cost sharing agreements among departments or researchers are acceptable techniques for distributing applications’ cost burdens.
Due to the shared nature of a computing cluster and the need to maximize its availability, users must not install any software on the cluster. Under no circumstances will software that may negatively impact the performance or stability of any component of the cluster be installed on the system.
If a user’s coursework or research requires the installation of additional ACC approved software, ECS’ cluster administration staff will evaluate the package(s) and if deemed stable, install the software upon request. If a package is determined to be unstable, incompatible or inappropriate for use on the College cluster, the requestor will be informed of the reasons for that decision.
Other Software Available:
Cluster’s ROCKS Bundle:
- Intel C/C++ Compilers 9.0-030
- Intel Fortran Compilers 9.0-031
- Intel Application Debugger 9.0
- Intel Cluster Math Kernel Library 8.0
- Intel MPI 2.0
- Intel Integrated Performance Primitives 4.1.3
- Intel Trace Analyzer 6.0.10
- Intel Trace Collector 6.0.1.0
- MPICH 1.2.7p1 compiled with Intel Compilers
- LAM 7.1.1 compiled with Intel Compilers
Mathematica
Cluster Backup
Each user will be assigned a “Home” folder within the system’s “Home” folder. Only files relating to jobs on the system will be stored in an individual’s “Home” folder. Students violating this policy will be subject to reprimand; continued violation of the policy could result in immediate loss of privileges on the system. Data placed in an individual’s “Home” folder will be backed-up nightly to the central tape-based data storage repository during the College’s scheduled back-up cycle. All files will be purged according to the “Account Procedures” delineated above.
Scheduling
Initially, the system will have fifty (50) concurrent licenses for MatLab use; this license count will be subject to annual reviews in July. As stated in the previous section, these licenses are for class related work only. Priority for cluster use will be given to classes that are in session; when no classes are in session, priority access will be assured for students registered for classes using the system. All class accounts will cease to be active NLT 30 days after the conclusion of each semester and after the completion of final exams.
During semesters, registered research users will be given access based on the availability of licenses; i.e., licenses not allocated to class enrollees. Only the number of remaining free concurrent licenses will limit access to the system. This category of users will be afforded priority access to the cluster two times annually: 1) after finals in the fall semester and before the start of spring semester, 2) after spring semester finals and before fall semester classes.
Sun Grid Engine is the scheduling software used by the system to schedule jobs. Observing the cluster utilization is open to anyone within the College of Engineering domain; utilization may be viewed at http://engrhpc.ecs.psu.edu/ganglia.
All users must abide by current scheduling policies. These policies are subject to change based on load, software changes, maintenance requirements and security issues.
Violations
Any violations of cluster use policy, College computing policies, or University computing policies may result in suspension of computing privileges or other disciplinary action.
Disclaimers
These policies are subject to change; all changes will be reviewed and approved by the College’s ACC membership. The latest policies are available online at the following web site: http://www.ecs.psu.edu/communications/policies/polmain.asp. Users of the College of Engineering’s high performance computing cluster are responsible for ensuring their utilization of College computing resources is in compliance with all College and University policies.
Last update: 11/3/2006
|
