A Workshop on Effective Approaches to

Campus Research Computing Cyberinfrastructure

April 25-27, 2006
Crystal Gateway Marriott Hotel
1700 Jefferson Davis Highway
Arlington, VA 22202

Home || Workshop Proposal || Agenda/Presentations || Roster || Final Report

Summary:

Historically, campuses have provided some measure of research computing capacity to scientists and engineers. With the advent of supercomputing centers and departmental facilities, campuses began to reduce their involvement in provisioning computing cycles to research communities. Now, a richer understanding of the realities of cyberinfrastructure reinforces the value in having greater enterprise-level research computing capabilities. Moreover, if those capabilities were deployed in a consistent manner, then efficiencies, greater user effectiveness, and the possibility for inter-institutional leverage could occur.

We propose a workshop, aimed primarily at major campus CTOs but with relevance to CIOs, to begin a community process for greater common understanding around the technology options and, to a lesser degree, the political and financial issues for campus research computing cyberinfrastructure. The workshop will be centered around coordinated presentations by several universities that are doing leading-edge deployments of research computing capabilities. Campuses will be selected to present an architecturally diverse set of approaches, including campus-wide Grids, clusters (e.g. Condor pools or Beowulf clusters), condos, etc.

In order for the workshop to have impact beyond the immediate participants, it is important that it provide outputs beyond a website of powerpoint presentations. To that end, we intend to deliver a whitepaper detailing the approaches and the comparative analyses that emerge from the workshop discussions. We also intend to create a list of issues for further consideration by NSF in its goal of promoting campus research cyberinfrastructure. Lastly, we intend to take advantage of the many outreach opportunities that Internet2 offers for dissemination of the white paper and workshop results.

Background and Planning:

A planning committee will be convened immediately to refine the initial agenda, develop a set of specific topics to be addressed during the correlated campus presentations, and review the outputs of the workshop. The committee will be chaired by Kevin Morooney, PSU, with representatives selected from among the following schools: Purdue, USC, Texas, UAB, U Va, Oklahoma, Georgetown, Great Plains-affiliated universities, SURA-affiliated NMI activities, Wisconsin, and perhaps others.

Workshop Goals and Outcomes:

The workshop is divided into two sections. The first, a higher-level part of the program will identify basic technical options and discuss motivations and issues for campuses to consider as they move forward. It will also map out and develop the sets of issues that will need to be worked in creating a true multi-campus sharing, from technical aspects to policy and financials factors.

The bulk of the workshop will consist of a set of coordinated presentations from major research universities that are or are implementing campus research computing cyberinfrastructure. The presentations will focus on at least the following sets of issues:

Technical architectures: There are several major alternatives for research computing cyberinfrastructure, including Condor pools, campus Grids, Beowulf clusters, BlueGene class systems, and condos. Campuses will be selected to present that cover this full spectrum of implementation choices. Discussion topics will include cost/performance issues, physical facilities issues, fit of architectures to application requirements, issues in linking resources across multiple campuses, etc.

Associated data storage issues: Many high-end computation problems require large amounts of data and perhaps intermediate storage. Locating the data, striping and other appropriate staging approaches can make considerable difference in performance, etc. and the campuses will be asked to present on the options they chosen, rationale, and levels of satisfaction with the option selected. Additionally, as the data collection capabilities of federally funded labs and centers improves, how are campuses helping to manage productive access to this data? Finally, new compliance requirements for security and privacy will affect approaches to managing the data.

Connectivity across campus and to external networks for these compute servers: For centralized servers, there are issues associated with provisioning outputs to the visualization engines and other post processing that are usually back in the scientists' laboratories. For external connectivity, advanced networks and direct lambda connections pose interesting options and security challenges.

Building a cooperative effort: Frequently, a campus will seek to leverage one or several computing grants given to individual research groups to create a campus computing coalition. Approaches to building these coalitions, in areas such as proper machine environments (especially for space and environmentally constrained departmental labs), greater computing capacity opportunities, and security need to be identified and verified. Management issues: The ongoing costs of maintaining these environments, particularly for specialized hardware or operating system/grid layer software, need to be allocated, either to users, funding agencies, or campus IT management.

Management must also address the periodic need to synchronously upgrade the computing service, regardless of whether the subsystems were acquired piecemeal over several years. And, there is the need to establish allocation of the computing resources themselves to the subscriber community.

Workshop:

The workshop is planned to be adjacent to the Internet2 member meeting, in order to minimize costs and maximize visibility. The first half day is targeted to both campus CIOs and CTOs, while the following day is aimed more at CTOs and those more directly charged with campus research computing support.

The proposed agenda may be modified by the planning committee or NSF input:

April 25 2-5 pm
Overview and Motivation -a set of campus CIOs who have decided to provision high performance computing support, why they did so, what are the issues and what are the results to date. As a panel, they then explore inter-institutional use cases and issues. The session would also include a detailed look at the rest of the workshop agenda, so that other CIOs will understand the nature of the issues around alternatives

April 26 noon-6 pm
Technical Issues - A set of generally technical presentations by campuses that have or are implementing campus computing cyberinfrastructure, including many of those on the planning committee. The presentations will be coordinated to address a common set of important issues (identified by the PC). The participants will be selected to insure a wide variety of technical approaches, ranging from Grids to clusters to condos to large iron to other choices. Possible analytic themes include development of facility, code and hardware issues, key library conversions to the platform, allocation strategies, upgrade options, etc.

Dinner with NSF OCI Speaker and perhaps some other luminary BOFs

April 27 8-10 am
What are the technical (and policy) issues in tying together campus computing cyberinfrastructure between campuses - A mix of presentation and discussion, Presentations may include projects such as the GridShib family of approaches, allocations and accounting, performance, etc.

10:30-noon
Extended discussion/panel on commonalities and differences among campus approaches to deploying research computing cyberinfrastructure and the reasons for them. Panel also identifies key issues to address to foster inter-institutional use.

Outputs:

It is important for the workshop to begin to inform a much larger process to follow, as well as reach a broader audience of campuses beyond those attending. To those ends, we propose the following deliverables:

A roadmap of Campus Issues - a report to NSF identifying the key themes that emerge from the workshop, including motivations, approaches, choices, and outcomes.

A set of Case Studies - a web site, maintained by Internet2, that contains the approaches of individual campuses, using a common topical presentation on the key issues and their reasons for the choices.

A list of potential leverage points - a report to NSF identifying possible opportunities for community action (e.g. shared libraries of key domain apps that have been ported to various architectures)

A Map of Inter-institutional issues - a report to NSF that describes the major challenges, and possible approaches, in the interrealm sharing of these resources

This workshop is supported with funding from Internet2, Pennsylvania State University,
and the NSF - Grant No. OCI-0627970. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation (NSF).