We were intrigued to read about a training roadmap for HPC that was presented during a session at the recent TeraGrid 2011 conference. The paper was co-authored by Scott Lathrop of the National Center for Supercomputing Applications and Mark Richards. We caught up with Richards and Lathrop to learn more.
iSGTW: Welcome, and thank you for agreeing to speak with us.
Mark Richards: Thank you for your interest in this project.
iSGTW: At TeraGrid 2011, you presented a training roadmap for HPC. What exactly is the training roadmap?
Richards: The training roadmap orients new HPC users to the world of high performance computing and helps them determine which skills they will need to develop, which tools they can utilize to improve their productivity, and which training resources are available to help them on their way. We provide a flowchart that outlines the different concepts and skills and tools that are crucial to success in the HPC community. The flowchart provides a broad overview of high performance computing. The nodes in the chart lead the user to short descriptions of key concepts and tools and provide links to resources where the user can find additional information. Arrows between the nodes suggest a natural ordering of the content.
iSGTW: What made you decide that this roadmap was necessary?
Richards: Many new HPC users are simply not prepared to effectively use the most advanced computational technologies. They aren't utilizing the right tools to maximize their own productivity, and the programs they write are making inefficient use of the supercomputers.
Most users of HPC resources are primarily interested in using technology to advance the frontiers of science and engineering. They aren't interested in computing for its own sake. Surveys have shown that the vast majority of HPC users have been formally educated as scientists or engineers rather than programmers. Their programming skills are self-taught. So on the one hand, we see a great need to change academic curricula to improve students' preparedness to effectively utilize supercomputing resources. Others are actively pursuing this goal.
Our efforts with this training roadmap are complementary. We are targeting those who are in a position to learn on their own. Many users are first exposed to HPC as graduate students when they no longer have the time or inclination to take many additional classes in computing. But they have strong technical backgrounds and have the capacity to learn the ins and outs of high performance computing. Quality training materials and productivity-enhancing tools exist, and many are freely available online. The key challenge is to help the users figure out where they should be looking for help. In many cases, users could easily find the help that they need if they only knew what keywords to type into a search engine. But they don't know what they don't know!
iSGTW: Do you think these types of users can necessarily tell if their research requires HPC as opposed to high-throughput computing?
Scott Lathrop: Part of the training that will benefit people is helping them with understanding the differences between HPC and HTC to help them select an appropriate computing environment. For some projects it's not an either-or decision, but rather a process of deciding how to balance the computing elements among HPC and HTC resources.
iSGTW: What are you hoping people will gain from the roadmap?
Richards: We hope to give new users a broad overview of the concepts, tools, and technologies of high performance computing. It's not possible for new users to learn everything all at once. Our roadmap will provide concise explanations of the most important concepts, at a level that a new user can read and understand in a few minutes. To get more detailed information, users can follow links to additional resources such as tutorials, articles, videos, and books.
We hope users will be able to determine for themselves which topics they need to tackle right away to solve their most immediate problems, and which items they can return to later as the need arises. But we want to orient them to the space early, so that when they encounter new challenges or questions down the road, they will recall that the roadmap addresses those issues and has links to helpful resources. For example, a new user may not need to use performance analysis tools such as TAU. But we want to alert even the newest users to the existence of such tools, so that they will know what to look for when they do run into performance issues. And our roadmap would suggest a progression of tools, say from gprof to IPM to TAU, with the goal of helping users identify and learn the simplest tool that meets their needs.
We also hope that the roadmap will have a social element to it. We invite users to add descriptions and resources according to their own expertise. Feedback from users will help us identify the best resources for each topic or tool.
iSGTW: Will the roadmap be a wiki?
Lathrop: The roadmap is an evolving set of resources that will expand in breadth as new techniques and methodologies emerge. The resources will expand as people develop new and improved training resources. The community will be key to this, and the plan is to use an environment (starting with a wiki) that allows the community to provide feedback, requests for other information, and content to enrich the available resources. We welcome the community to contribute to the evolution and improvement of the resources.
iSGTW: As I understand it, this roadmap is just a small part of a much larger project you're working on. Can you tell us a little bit about that?
Richards: The training roadmap is part of a broader effort to improve education and training in the HPC community (See hpcuniversity.org). We are also engaged in outreach activities to help increase understanding of how HPC benefits society and how people can become part of the community. Resources for students include weekly challenge problems; links to internships, fellowships and job opportunities; and information about HPC programs at the high school and university level. Educator resources include descriptions of HPC core competencies and curricula. A calendar of events includes information about upcoming conferences, training events, and deadlines. We also provide a forum for users to discuss HPC issues.
iSGTW: Is the HPC University database of resources maintained so as to prevent dead links, and if so, by who?
Lathrop: While the site is hosted at Shodor (www.shodor.org), it is being maintained by multiple organizations through multiple funding sources. Content is being provided by the global HPC community. We welcome support from organizations who want to see the resource grow and expand to best serve the community.
iSGTW: When do you expect that the roadmap will launch online, and where will it be hosted?
Lathrop: The roadmap is expected to be released and made publicly available by September of 2011. It will be hosted at www.hpcuniversity.org and we hope others will want to link to and/or mirror the resource.