Enabling Extreme Computing
Grid Computing Simplified
esearchers have made significant science advances using Globus as a gateway to computing resources, with major projects underway in fields from high-energy physics to atmospheric modeling. By providing consistent and uniform access to Grid computing, much like the power grid provides electricity, Globus has altered the way scientists and engineers use computers. In addition, SDSC researchers have developed Web-based tools that simplify Grid access by interacting transparently with Globus, allowing scientists to concentrate on their research rather than the details of the supercomputer.
|Figure 1. Gravity Waves
This visualization of gravitational waves, computed in accordance with Einstein’s theory of General Relativity, was generated using Globus with mpichG2 and Cactus software. Visualization created by Werner Benger, NCSA/AEI Potsdam/Wash U/ZIB visualization team, and provided courtesy of the Max Planck Institute for Gravitational Physics, and Konrad Zuse Center for Scientific Computing.
Globus has become an essential piece of infrastructure in distributed high-performance computing by providing the software and tools needed to integrate geographically distributed scientific instruments, databases, and tools for scientific visualization. Over the past few years, Globus has been deployed at more than 100 sites worldwide, including the PACI centers.
"We’ve built a foundation that’s become the de facto standard," said Ian Foster of Argonne National Laboratory and the University of Chicago. "At this point, there seems to be a broad acceptance of Globus technology for high-end systems." Foster and Carl Kesselman of the Information Sciences Institute (ISI) at the University of Southern California, have been collaborating on the project since 1995. Globus is a joint project of ISI and Argonne National Laboratory and is a focus of development at both NPACI and the National Computational Science Alliance.
One scientific development that will make the most of Globus’ features is the Grid Physics Network (GriPhyN), led by Foster and Paul Avery of the University of Florida. The GriPhyN collaboration, which consists of a team of experimental physicists and information technology researchers, aims to implement the first petabyte-scale computational environment, the Petascale Virtual Data Grid (PVDG).
The PVDG will provide computing resources for researchers working on four experiments examining the fundamental forces of nature and the structure of the universe. The Compact Muon Solenoid and A Toroidal LHC ApparatuS (ATLAS) experiments at the Large Hadron Collider at CERN will search for the origins of mass and probe matter at the smallest length scales; the Laser Interferometer Gravitational-wave Observatory will detect the gravitational waves of pulsars, supernovae and in-spiraling binary stars; and the Sloan Digital Sky Survey will carry out an automated sky survey enabling systematic studies of stars, galaxies, nebula, and large-scale structure.
The data sets generated by the experiments are expected to grow from 100 terabytes to a 100 petabytes over the next decade. In addition, GriPhyN is expected to perform computations that require more than 120 trillion floating-point operations per second. Such a massive computing effort will require linking thousands of computers worldwide, said Miron Livny of the University of Wisconsin. Globus, in tandem with other automatic resource brokering software, will provide a set of low-level protocols for resource access and connectivity for those computers.
"Globus is one part of a virtual toolkit of different technologies that will be used for the project," said Livny. Other NPACI technologies forming part of the GriPhyN environment are NPACI Rocks for managing linux clusters and the SDSC Storage Resource Broker for managing the large scale data sets.
While Livny said the project is still in a relatively early planning phase, GriPhyN researchers recently did a successful trial run on Alliance resources using data from CERN’s Compact Muon Solenoid experiment. The group used Globus, in conjunction with Condor, to move data between computing resources at Caltech, the University of Wisconsin and National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign.
Click to view larger image.
|Figure 2. How Does GridPort Work?
Enabling Extreme Computing
Another data-intensive project which tests Globus’ ability to handle extreme computing demands has been the Cactus project. The modular, open-source problem-solving environment for scientific and engineering problems, was originally designed to tackle some of the most complex problems in astrophysics, including Einstein’s equations for colliding neutron stars, black holes, and the formation of singularities.
Researchers from the Cactus group recently collaborated on a distributed run using Globus with mpich-G2 and Cactus software to prove that massive computing problems could be spread among multiple supercomputers. The simulation of gravitational waves formally known as Teukolsky waves lasted between four to eight hours and used 512 processors on three SGI Origin systems at NCSA and 1,024 processors on NPACI’s Blue Horizon at SDSC (Figure 1).
"We ran a large test case as a single Globus job and it ran like a champ. Best of all, even though the code had been scaled up to run on more than 1,500 processors, it executed at better than 70 percent efficiency," said John Towns, Division Director of Scientific Computing at NCSA.
Thomas Dramlitsch of the Max Planck Institute for Gravitational Physics said, "This run demonstrated that this kind of resource coalescing is possible. Furthermore, we achieved something more: we could show an adequate scaling–up to 70 percent. Older experiments had only two supercomputers involved and the scaling was below 50 percent."
Grid Computing Simplified
However, working with Globus still requires technical expertise and a budget that may be out of reach for some research groups. To address those issues, NPACI has created services such as the NPACI HotPage and GridPort, which allow Web access to the PACI Grid resources, yet hide complex interactions with Grid software. Globus toolkits such as Resource Management, Globus I/O, and Grid Security Insfrastructure are used extensively by GridPort.
The original version of GridPort broke ground by providing a software development toolkit of standard, portable technologies on top of Grid software such as Globus that developers could use to create scientific application portals. These portals could then be used to securely access high-end computing resources from any Web-connected device, including wireless handhelds. Last November, GridPort’s "HPC anywhere" capabilities were demonstrated at SC2000.
The GridPort development team, led by Mary Thomas at SDSC, has since taken the idea of portal development one step further. The group recently released a beta version of the GridPort Client Toolkit that allows scientists who know how to build Web pages to create customized application portals that access the PACI Grid. Researchers need only to incorporate a few extra lines of HTML into a Web page to set up a Web site running on any server, anywhere in the world, that can communicate, via Globus and other software, with the PACI Grid (Figure 2).
"This capability has been a long-term design goal of the NPACI portal effort," said Thomas, manager of the SDSC Computational Science Portals group. "It will allow individual scientists to build simple portals that take advantage of existing infrastructure without having to invest large amounts of time or funding. These portals allow scientists to concentrate on research rather than the details of the supercomputer."
Current features allow users to access services such as logging in, uploading files, and submitting jobs, and more extensions that are planned for the future. "Now, essentially all it takes to create a customized application portal to GridPort is to incorporate a few extra lines of HTML into a Web page," said Maytal Dahan, a portal developer.
The GridPort Client Toolkit was demonstrated in a tutorial at the NPACI All-Hands Meeting 2001. Students were given NPACI portal accounts; they then downloaded an example set of HTML pages; finally, they installed the pages either onto a Web server or ran them locally. With very few modifications to the downloaded Web pages, the students had an instance of a simple, working portal.
Research scientist Don Sutton successfully demonstrated the ease of use of these tools, which he recently incorporated into the existing Basins, Bays, and Estuaries Project (BBE), part of NPACI’s Earth Systems Science thrust area. Sutton migrated an atmospheric modeling code, which had been running on local SDSC workstations, to HPC resources and connected the code to the BBE portal by adding security and job submission capabilities via the GridPort Client Toolkit.
"Grid technologies, such as Globus, have made it possible for application developers to redefine how users interface with NPACI resources," said Jay Boisseau, former SDSC associate director for Scientific Computing and now director of the Texas Advanced Computing Center at the University of Texas, Austin. "Globus, in particular, allows GridPort developers to extend functionality via the Web, which is a tremendous tool for doing computational science." –CF
Argonne National Laboratory and the University of Chicago
Information Sciences Institute, University of Southern California
I. Foster, C. Kesselman, S. Tuecke, (2001), "The Anatomy of the Grid: Enabling Scalable Virtual Organizations," to be published in Intl. J. Supercomputer Applications.