Understanding StarLight: An Interview with Joe Mambretti
Reprinted from HPCwire, the high performance computing e-magazine Nov. 14, 2001: Vol. 8, No. 2
L I V E w i r e from H P C w i r e
Brought to you Live from SC2001, DENVER, COLORADO
UNDERSTANDING StarLight: AN INTERVIEW WITH JOE MAMBRETTI
by Alan Beck, editor-in-chief, LIVEwire
Denver, Colorado -- One of the principal topic areas addressed in this year's MasterWorks track is HPC Computing Infrastructure; within this focus is a discussion of "StarLight: Optical Switching for the Global Grid." To learn more, HPCwire interviewed Joe Mambretti, Director, International Center for Advanced Internet Research, http://www.icair.org.
HPCwire: Please describe the background, funding, specific resources, and mission of StarLight.
MAMBRETTI: StarLight/STAR TAP (Science, Technology and Research Transit Access Point): STAR TAP, a National Science Foundation-funded project of the Electronic Visualization Laboratory of UIC is the world's only connection point for advanced networks world wide. It connects advanced networks form all the major world economies. StarLight is the Optical next generation STAR TAP.
StarLight is being developed by the Electronic Visualization Laboratory at the University of Illinois at Chicago, the International Center for Advanced Internet Research at Northwestern, and the Math and Computer Science Division at Argonne National Laboratory.
StarLight combines advanced optical infrastructure and leading edge facilities as a foundation for network services optimized for global high-performance applications. A key focus will be on interlinking high performance computer clusters world-wide, to support e-science, which tends to be data and computationally intense.
We aim to create a global high performance computer back-plane The National Science Foundation provides major support for this project (http://www.startap.net/starlight). Additional resources are provided by the multiple partners who are involved in this project.
HPCwire: What is your specific role in StarLight?
MAMBRETTI: The StarLight consortium is designing, developing, and managing this facility.
HPCwire: How may an individual, enterprise, or institution make use of its facilities and/or personnel?
MAMBRETTI: Individual researchers will participate by using StarLight for advanced, high performance global applications such as those being developed for iGRID 2002 in Amsterdam (see www.startap.net). Institutions, especially international advanced networks will participate by linking to this facility. For example:
NetherLight: This project, which is being funded by the government of the Netherlands through the GigaPort Network, is creating the world's first trans-Atlantic wavelength (2.5 Gbps) devoted to research. The link will connect StarLight and a science center in Amsterdam. GigaPort is a joint project of the Dutch government, trade and industry, educational institutions and research institutes. The aim of GigaPort is to provide the Netherlands with advanced, innovative technology. The GigaPort Network is one of the world's leading research networks (GigaPort is implemented under the authority of the GigaPort Steering Committee. GigaPort Network is realised by SURFnet; GigaPort Applications by the Telematica Instituut. (http://www.gigaport.nl)
HPCwire: How mature is optical switching technology, and what will be the next steps in its development?
MAMBRETTI: The basic technology used, DWDM, is mature, having been used in long-haul networks for over 12 years. However, this technology is now migrating to regions, to metro areas, and even to the enterprise. The next step is to optimize it for individual applications. One of the goals of this consortium is to allow for dynamic lambda switching by individual applications.
HPCwire: How will the roles of optical and traditional switching evolve in grid-intensive computing?
MAMBRETTI: Essentially, new technologies allow for optical networks to be used as distributed back-planes for high performance computers. StarLight will be implementing this technique for clusters world-wide. In the US, a related $53M NSF project, the TeraGrid, has been announced. StarLight will be a node on the TeraGrid, and will offer to provide access for the international community. This TeraGrid facility will initially link computers at four locations, the National Center for Supercomputing Applications, Caltech, the San Diego Supercomputing Center at the University of California in San Diego, and Argonne National Laboratory.
HPCwire: What are the principal challenges that must be overcome?
MAMBRETTI: There are a variety of issues to be addressed. For example, there are multiple contending architectures for L3 signaling methods. The approach of StarLight will be to develop methods that will optimize a optical networking architecture for global e-science.
HPCwire: Are wholly new and radical technologies needed for a truly effective synthesis?
MAMBRETTI: The goal will be to utilize some existing techniques, protocols, technologies, and architecture, to extend some these, and to create others that are totally new.
HPCwire: Are performance comparisons of grid-intensive computing platforms to proprietary vector-based and/or multithreaded machines still relevant? Why or why not?
MAMBRETTI: The emerging structures can accommodate both types. The question applies more to computing architecture. However, one implication is that a system of inexpensive distributed clusters may offer significantly more processing power than any individual proprietary system.
HPCwire: What part will optical switching play in the ubiquitous Internet envisioned by Cal(IT)2 and others?
MAMBRETTI: It's a core technology that serves as a foundation for generalized systems.
HPCwire: What contributions can be expected from private industry? From governmental and nonprofit entities?
MAMBRETTI: All of these types of efforts are very large scale and benefit from partnerships among many types of organizations.
HPCwire: Is there anything else you would like our readers to know?
MAMBRETTI: StarLight is a unique facility that will provide for powerful new capabilities will lead to a new set of advanced applications. Readers should note the projects related to the iGRID efforts for some examples.