Many graduate students work on computationally intensive projects and can benefit from the use of high performance computing methods. There are three main kinds of high performance computing resources available to students working on their third year papers or on their dissertations:
Linux Hawk Cluster
The department has a Linux computing cluster, reserved for graduate students, named hawk, which has 144 compute nodes in total. The key advantage of using the cluster is that students can take advantage of parallel processing, which allows a program to run on more than one compute node at a time. Students can access up to 36 nodes at a time. In their fourth and fifth years (in preparation for the job market), students can access up to 60 nodes. The hawk cluster runs the Linux operating system and has available both Fortran (Intel and GNU) and C compilers. The cluster also has available a parallel optimizer called HOPSPACK. To get an account on hawk, contact Professor Todd.
For information on the Linux system editors, a recommended book is The Ultimate Guide to VI and EX Text Editors (Hewlett-Packard).
For information on the HOPSPACK optimizer, see Sandia National Laboratories' HOPSPACK page.
For information on learning MPI (Message Passing Interface language), a recommended book is MPI-The Complete Reference, Volumes I and II by William Gropp, Jack Dongarra, Steven Huss-Lederman, Andrew Lumsdaine, Ewing Lusk, Bill Nitzberg, Steve Otto, William Saphir, Marc Snir, and David Walker.
Tesla GPU/Statistical Server
For students who only need a single-node server to run jobs in R or Stata, the School of Arts & Sciences provides the Tesla server, which runs in Linux. In addition, Tesla contains one Tesla M2075 GPU card (its namesake) available to faculty and students whose work lends itself to GPU processing. The server supports development is C and Fortran (through the PGI compilers) in CUDA 2.0.
The Extreme Science and Engineering Discovery Environment (XSEDE) is the most advanced, powerful and robust collection of integrated digital resources and services in the world and is supported by the National Science Foundation. Students, with the support of a faculty member, can apply to use XSEDE resources. The application process is competitive, although it is rare for applications to get turned down.
The XSEDE environment is similar to the environment on the hawk cluster (see above) in that it also runs a Linux system platform and the process by which jobs are submitted is similar to the queuing process used on hawk. XSEDE has available a variety of software packages, including Fortran and C compilers with MPI capability and, on some computer systems, MATLAB. HOPSPACK optimizer can also be installed if not already done so.
By using XSEDE instead of the hawk cluster, students can access a higher level of resources (more nodes, faster CPUs with very high speed connectivity). It is often desirable to first develop programs on the departmental cluster (hawk) and then, when the program is running, transfer it to one of the supercomputers accessable from XSEDE. Some of the supercomputers can be accessed using SSH protocol, while others require Globus (a more secure way of accessing it).
The Application Process:
To apply to use XSEDE resources, students need a research abstract describing the research and how the computing resources will be used. A faculty member must sign on in support of your project as a co-applicant and the application also requires the CV of the faculty member. The application needs to be submitted through the XSEDE User Portal.
There are two stages to the application process. In the first stage, the student signs up for a “start-up allocation” which takes about two weeks to get approved. This provides the student with an initial set of resources for about two months for testing the program. This may be all that the student needs. However, if more resources are needed or the initial set is needed over a longer period of time, then the student needs to apply for a “research allocation” which takes a little longer to get approved. Usually, the resources are given for one year at a time and students have to reapply to use for another year. The entire application process is through the webpage.
Gavin Burris is Penn’s XSEDE representative. He can help determine which XSEDE resources are most appropriate.
Graduate Data Analysis Lab (GDAL)
The Graduate Data Analysis Laboratory (GDAL) has software for performing quantitative analysis. GDAL has several Windows computers and is located in the McNeil Building, room 303. All of the computers connect to a Windows terminal server with 48 CPU cores and 192GB of RAM. This powerful server allows the students to run single instances of jobs where individual PCs often required segmenting the jobs into several parts.
Graduate students in Social Sciences Departments and Centers are eligible to use GDAL. Send requests about the lab to: LABS-LSP@ssc.upenn.edu.
Other Related Computing Resources
AirPennNet — Secure Wireless Network
The University provides a secure wireless network to all students, faculty, and staff. AirPennNet utilizes the 802.1X protocol to provide an encrypted wireless connection and works from many computing and mobile platforms.
More information about connecting to AirPennNet is available at Penn Computing's AirPennNet page.
Google Apps for Education — Gmail and more
The School of Arts & Sciences is currently providing Google Apps for Education as an optional email service for SAS students. The service includes a customized version of the popular Gmail service, in addition to several other Google Apps (Calendar, Docs/Drive, etc.) with special privacy assurances from Google.
More information about Google@SAS is available at SAS Computing's Google@SAS FAQ page.