|
|
|
|
Link to slides used for this paper as presented at the conference. They are in Acrobat PDF format.
The Ohio Supercomputer Center has a major mandate from the state of Ohio to provide high performance computing facilities and services to Ohio academicians and commercial clients. The Center's organization includes the HPC branch, comprising the groups who handle the essential day-to-day tasks that support the Center and its users. The Business and Administration Group provides financial and personnel support, while the Software Group licenses and installs third-party and public domain software applications and libraries. Ohio researchers can call on the expertise of the Science and Collaboration Group, which also develops its own special research projects. The Systems Support Group administers the HPC facilities, staff systems, and network and has played a key role in the transitions to new machines during the ten years the Center has functioned. Finally, HPC's Science and Technology Support Group provides technical assistance, HPC training, and technical documentation to the user community and collaborates with scientists and engineers through Ohio.
The Center maintains a number of HPC systems, including a CRAY T90, CRAY T3E, CRAY J90, SGI Power Challenge, and an IBM RS/6000 SP. The table below highlights the number of processors and memory configurations.
| System | Processors | Memory |
| CRAY T90 |
4 |
128 Mw |
| CRAY T3E |
128 |
16Mw/pe |
| CRAY J90 |
16 |
|
| SGI PC |
16 |
2 Gb |
| IBM SP |
8 |
256 Mb/p(5) 128 Mb/p(3) |
The SGI/Cray products are the Center's production systems, used by a large number of users. The most utilized system is the CRAY T90, followed by the CRAY T3E. In recent months, the CRAY T90 has been running very near 100% capacity.
While the Center is centrally located in Columbus, Ohio, the user community is geographically dispersed throughout the state. The majority of the user community includes faculty and graduate students from more than 20 universities in Ohio. The majority of scientists and engineers are represented by the National Science Foundation (NSF) science codes Mathematics, Astronomy, Physics, Chemistry, and Materials Research. Approximately 25% percent of the user community runs third-party applications, such as ANSYS and LS-DYNA. The rest of the community develop and execute their own computer codes. The focus of HPC training offerings at the Ohio Supercomputer Center are developed and designed for vector and parallel programmers to efficiently and effectively execute computer codes on the CRAY T90 and CRAY T3E.
The Ohio Supercomputer Center offers an array of training courses for the HPC users, in both the academic and Department of Defense communities. Short courses lasting from 1 to 3 days offer applied techniques on HPC systems. They cover operating system techniques, programming languages, parallel programming environments, system utilities, performance tools, and debuggers. Two recently developed courses are "Using the CRAY T90 at OSC" (T94) and "Using the CRAY T3E at OSC" (T3E). Both follow a similar outline and correspond to nomenclature found in technical documentation. Broad topics include
While most topical areas may be apparent, a few need further clarification. For the T3E course, the operating system overview describes UNICOS/mk, identifies types of processor elements, and demonstrates commands such as getpid, malloc, open, and suspend. The operation environment is the interactive and batch environment established by the site and includes access information and commands for managing processes and local environments (i.e., shells, modules, file migration, NQE commands, batch queues, and temporary disk space). The programming environment covers the compiling systems, debuggers, and performance analysis tools. Issues are typically workarounds for known bugs for the current state of the system and change with new system and software releases. Before the installation of MPT 1.2, a known bug with the MPI_Bcast routine was discussed. Although it requires a rare combination of features, they were known to be used by the scalapack library from NetLib[1].
Training materials originate from a variety of electronic sources -- from hand-coded html to the use of commercial off-the-shelf packages, such as Microsoft's PowerPoint and Claris Works. Whatever the origin of the materials, they are converted for Web publishing either as text with images or images only. For presentation and capturing purposes the materials are separated into files so that each file includes one page. A style guide is used to number the pages and select appropriate font sizes for presentation. Instructor-based presentations combine the slides with a navigational frame that includes hyperlinks for the table of contents, section index, and previous and next slides. The navigational frame is generated through the use of perl scripts that track indices and page numbers.
Training materials are developed by the instructors and based on their experiences and knowledge. Each handout provides key points of interest and is an indicator on which the instructor can elaborate for the given skill levels and interests of the audience. Most of the information is transferred through what the instructor says. Capturing the audio from the lectures and synchronizing it with the course materials for delivery on the Web provide a mechanism for recycling the presentation to reach a wider audience.
To quickly reach the widest number of geographically distributed users without expending a large number of resources, the Center uses the Web Lecture System [2] for real-time capturing of training workshops. WLS combines existing technologies -- Web browsers, HTTP servers, HTML documents, CGI programs, Javascript, Java Applets, and RealAudio -- to make these presentations available via the Internet. Because the resulting product does not require a high-bandwidth solution, the presentations can be viewed and heard by the widest possible audience. RealAudio uses streaming technologies and can offer audio without stressing the network. WLS provides the interface for automatically synchronizing the audio and managing the materials. Materials are prepared and loaded into the WLS system, one section per lecture on a separate PC. We have found presentation and synchronization are best performed by separate individuals. The instructor presents the materials, and an operator records and synchronizes the audio. The instructor and students view the materials via a frames-capable Web browser, and the operator views the content pages within WLS. Typically, the operator is an alternate instructor for the given training course. These operations are interchangeable, so both instructor and operator are knowledgeable in content and synchronization methods. WLS provides the GUI for easily identifying the beginning and end times associated with each slide. To ensure a high quality sound at optimal levels, a wireless microphone is used for recording. The figure below illustrates the capturing set-up.
WLS provides three viewing schemes for Internet users. One is an automatic method for viewing the materials and hearing the instructor and students. When the automatic method is used, the observer views the slide and hears the audio. The slides are shifted automatically according to the timestamps in the synchronization file, as if the instructor were changing the slides during the presentation. Observers may also choose a manual method or a printable method. The manual method allows the observer to listen or not listen to a particular audio track for a given slide. For the printable method, WLS combines each of the separate files for a given lecture into one file that the observer may easily print.
Once the presentations are recorded, the files are moved to the Technical Information Web Site (http://oscinfo.osc.edu/) and placed in the training section.
We are pleasantly surprised with the functionality of the WLS system and the ability to record and synchronize audio files with slides in real-time and, in a fairly short time frame, provide the captured materials on the Web. The purpose of these observations is to share the lessons learned during our early attempts at using the system in an effort to ease the process if you so choose to implement the product.
While WLS provides most of the management of the HTML and sound files, it is often useful to understand the underlying structure, especially, when moving files between Web servers. Since the encoded synchronization files associated with the RealAudio server require absolute path names, some editing and RealAudio utilities are required to move to a different Web Server. WLS and RealAudio generate a number of files. WLS separates functions through frames and RealAudio requires both metafiles and sound files. Metafiles are used to point to sound files, map corresponding uniform resource locators (URLs), and if necessary, denote beginning and ending time stamps. As many as six files can be associated with a given page.
When users shift between automatic and manual methods, and when they move back or forth through single pages or groups of pages, navigational frames can be nested. If users frequently move back and forth in the material, users may notice a lag time between audio and slides.
When playing, the RealAudio Player displays timing information. While the RealAudio player offers streaming technology for audio, the initial loading of long audio files can require enough time that the observer assumes a problem or becomes impatient and quits. For these reasons, we indicate the length for each lecture; this way, the user, depending on capabilities and time, may choose whether to load the sound file(s). In this case, a lecture is generally a functional section of the presentation.
Although most of the transferred information is through the instructor's voice, delivery on the Web can be less than engaging for the learner. With the addition of streaming video from RealNetworks, a natural transition is to synchronize the video including audio with the training materials. Video provides a visual cue for the learner and adds another media element.
Real-time capturing of training presentations is a step toward independent, interactive, Web-based modules that focus on the learner. It provides a vehicle for improving the quality of materials by closely evaluating presentation content, assessing needs, identifying self-contained chunks of information, and articulating outcomes.
In the future, the Center plans to offer synchronous HPC training courses over the Web using Tango[3], developed at Syracuse University, and to develop asynchronous modules for training delivered on the Web with multiple media elements. A team is forming to understand instructional design methodologies for multimedia and develop modules using existing captured materials.
The Ohio Supercomputer Center thanks Dr. Mladen Vouk and Dr. Richard Klevans at North Carolina State University for their demonstration and installation assistance with WLS. Initial implementation would not have been possible without the experience gained at the Aeronautical System Center Major Shared Resource Center of the Department of Defense Modernization Program.
http://trurl.npac.syr.edu/tango/papers/tangowp.html
Leslie Southern has been at the Ohio Supercomputer Center almost from its inception, ten years ago. Hired as a consultant, Southern is now the Science and Technology Support Lead for the Center's HPC division. Her responsibilities range from coordinating consultation schedules, developing and teaching OSC's workshops, through working on the Center's annual two-week program for high school students. Lately, Southern has added training course and materials development for the Department of Defense Modernization program to her list of responsibilities. In addition, she supports the Center's collaborations with the National Science Computational Science Alliance Partners for Advanced Computational Services. Before joining the Center's staff, Southern worked at Battelle Memorial Institute on the Office of Nuclear Waste Isolation project.