3.09.05 -- “In the past,” says Andrew Chien, “there has been a huge gulf between running applications and exercising low-level control of the network. It has been almost impossible for anyone to control the net.
What Chien is alluding to is a demonstration done recently at the OptIPuter All-Hands Meeting moving toward user control of the net. The OptIPuter, so named for its use of Optical networking, Internet Protocol, computer storage, processing and visualization technologies, is an emerging infrastructure that tightly couples computational resources over dedicated optical networks using the Internet Protocol.
What’s exciting – and revolutionary – about the OptIPuter project is its promise of providing applications-driven, configurable optical networking to end users. This project is breaking new ground, according to Chien, because the traditional model is that the network is an infrastructure controlled by carriers, not end users.
A first tangible step towards user control was a so-called “three-layer” demonstration. These “layers” refer to a series of software components, which can be conceptualized as layered on a “stack.” Such software stacks are sometimes called “middleware.”
This demo linked three layers of what will eventually be a more complete five-layer middleware software stack. This three-layer stack combines (1) the ability to visualize extremely high-resolution data with (2) the ability to configure a “distributed virtual computer” (DVC) and (3) the use of novel data transport protocols. All three layers represent new research and development funded by the OptIPuter project. This approach allows applications to control their own resources as they are used, making adjustments moment by moment as required. Resources were aggregated from visualization and storage clusters at the National Center for Microscopy and Imaging Research (NCMIR) at UCSD (led by Mark Ellisman), the University of Illinois at Chicago, and the University of Amsterdam.
Jason Leigh, UIC, oversees the visualization team. Chien, a professor in the Computer Science and Engineering department and director of the Center for Networked Systems at UCSD, leads the DVC team and the protocols team, which includes Bob Grossman and Leigh at UIC and Joe Bannister at USC/Information Sciences Institute.
In the demonstration, JuxtaView, a 2-D image visualization system, was launched by the DVC to display a high-resolution image of a rat brain on a tiled display connected to the visualization cluster at NCMIR. JuxtaView exploits LambdaRAM, an efficient data service based on pre-fetching techniques, to retrieve raw data from remote storage servers at UIC and the University of Amsterdam through dedicated 10-Gbps CAVEwave and transatlantic links.
Several OptIPuter transport protocols can be used by LambdaRAM for remote data retrieval under a shared communication framework, called Globus XIO, developed by OptIPuter team member Carl Kesselman, USC/ISI, and his co-workers. XIO enables convenient selection and easy use of these protocols through a unified applications program interface.
The demo was supported by five research groups in Amsterdam, Chicago, and San Diego, requiring additional integration “on the sociological front” among the infrastructure and information technology staff.
A five-layer demo, reaching down in the stack to include user/software control of lightpaths and up to include applications in neuroscience and geophysics, is planned for the biannual iGRID meeting at the end of September. It is expected to feature neuroscience remote data access and display at the top level of the stack supported by OptIPuter distributed optical backplane software (PIN/ODIN) at the bottom level, extending the network control directly to lightpaths within the optical network by the end user at both ends of the stack. This software and these protocols are being developed by UIC and the International Center for Advanced Internet Research at Northwestern University, another OptIPuter partner.
“We’re architecting a framework,” says Chien, “that allows end users to bring together the right Grid resources and optimize networking support for their applications transparently. They get to make the best use of the technology without fussing with low-level details of networking technology or having to make changes to their applications. They’re justifiably very excited about this.”
It’s important to put the progress of research on configurable optical networks in the proper context. Research on dynamic optical networking with distributed control has been conducted on advanced metro area testbeds, such as OMNInet, which has been used to develop experimental protocols in support of the OptIPuter. However, the OptIPuter project is developing the middleware software to include many more features.
“People doing this were considered the lunatic fringe at the beginning of this project,” says Chien, “but now a broad community is tracking closely what we’re doing. That the National LambdaRail (NLR) is becoming a reality is proof positive of the importance of this project and advanced optical networking in general.”
“Other countries are certainly ahead of us in the game of national optical networks,” say Chien, “but NLR shows that that the U.S. is increasing activity rapidly. Though many don’t realize it yet, this community is developing critical mass and will have a profound impact.” Facilities such as the NLR are key resources for advanced distributed environments such as the OptIPuter.
Increasingly, the international advanced networking community is addressing the need for a global optical infrastructure. A quick perusal of the growth in the advanced optical networking community is reflected in increasing attendance at the annual Global Lambda Integrated Facility Workshop. According to Tom DeFanti, co-PI on the OptIPuter project and director of the Electronic Visualization Laboratory at UIC, attendance grew each year from 15 in 2001, when the meeting began, to 25 to 40 to 60 at last year’s meeting. Some 250 attendees are expected when it convenes this fall in tandem with the iGRID meeting at UCSD’s new Calit2 building.
For those wanting to take advantage of OptIPuter research and development, next year the project will make its software more generally available on the model of the SDSC Rocks software for cluster management, a project led by Phil Papadopoulos who is also co-PI on the OptIPuter project.
Special thanks to Nut Taesombut, a graduate student in Andrew Chien’s lab, for assistance in preparing this article.