The Quarterly Publication of the Numerical Aerospace Simulation Systems Division

NAS researchers are developing atomicscale transistors to enable future

microelectronics. Page 10

IPG Infrastructure Nears Completion Ñ 18

NREN Drives Global Deployment of Multicast Protocol Ñ 7

CART3D: A New Generation of Grids Ñ 4

CFD Supports Wind Turbine Research Ñ 14

Vol. 1, No. 3 • Summer 2000

Executive Editors: Bill FeiereisenJohn Ziebarth

Editor: Nicholas A. Veronico

Writers: Holly A. AmundsonToshishige Yamada

Graphic Designer: Sue Kim

Illustrator: Gail Felchle

Photographer: Michael Boswell

Contact the Editor:Nicholas A. VeronicoGridpointsM/S 258-6Moffett Field, CA 94035-1000Tel. (650) 604-3046Fax (650) 604-4377nveronico@mail.arc.nasa.gov

Gridpoints is a quarterly publication of the NASSystems Division at NASA Ames ResearchCenter. The division’s staff includes governmentemployees and contractors from Cray ResearchInc., Advanced Management Technology Inc.,Computer Sciences Corp., Eloret, RaytheonCompany, and SGI.

Gridpoints acknowledges that some words, modelnames and designations mentioned herein are theproperty of the trademark holder. Gridpoints usesthem for identification purposes only.

The staff of Gridpoints would like to thank thefollowing people for their assistance in preparingthis publication: Michael Aftosmis, DickAnderson, Bill Arasin, Michael Boswell, SamsonCheung, Bob Ciotti, Jill Dunbar, Earl Duque,Chris Gong, Bryan Green, Jennifer Goudey, T.R.Govindan, Chris Henze, Tom Hinke, Vee Hirsch,Amara de Keczer, Tony Keefe, Chris Kleiber, BobKufeld, Dochan Kwak, Eric Langhirt, TammyLeal, Bonita McPherson, George Makatura,Yvonne Malloy, Donovan Mathias, MeyyaMeyappan, Gina Morello, George Myers, ToddNovak, Jeff Onufer, Marcia Redmond, ScottSchreck, Ryan Spaulding, Roger Strawn, ValerieSteele, Ron Sykora, Jim Taft, Leigh Ann Tanner,Bill Thigpen, Richard Thompson, Ray Turney,Judith Utley, Rob Van Der Wijngaart, CliffWilliams, Michael Wilson, Mindy Wilson,Toshishige Yamada, Jerry Yan, Maurice Yarrow

Visit us on the web at:http://www.nas.nasa.gov/gridpoints

On The Cover:

Subscribe to GridpointsSubscriptions to Gridpoints are complementary.Please e-mail your name, company name, title,address, and zip code to:

gridpoints@nas.nasa.gov

Gate

Source Drain

Atomic: < 10-9 m (top)The Smallest Nanoelectronics: AtomicDevices with Precise StructuresTransistor miniaturization is reaching its limit. Atomicscale transistors are being developed to enable futuremicroelectronics.Toshishige Yamada

CFD Supports WindTurbine ResearchNAS provides the National Renewable EnergyLaboratory with computational resources to helpimprove next-generation wind turbine designs.Holly A. Amundson

The Quarterly Publication of the Numerical Aerospace Simulation Systems Division

4

18

7

NASA Research andEducation Network DrivesGlobal Deployment ofMulticast ProtocolMulticast will speed up the Internet by en-abling a single data packet to be distributedto multiple Internet users simultaneously.Hugh LaMaster and Kenneth Freeman

10

Network B

Router Router

Network A

Router

Router

RendezvousPoints

Network C

Router

RouterRendezvous

Points

RendezvousPoints

CART3D: A New Generationof Grids3-D grid software automatically generatesvehicle simulations from complex geometries.Holly A. Amundson

14

NAS researcher Toshishige Yamada has conducted experiments with atomically precise structures usinga substrate atomic lattice, floating of atoms above the substrate, and chemical bonding of atoms to thesubstrate. Yamada’s research will enable future atomic scale devices. See page 10 for the full story.

Infrastructure For NASA’sInformation Power Grid NearsCompletionMore than a dozen components of the IPG are beingreadied for roll-out this September.Nicholas A. Veronico

Features

NAS nanotechnology researcher Deepak Srivastavachaired the conference “Carbon Nanotubes: Advances

in Cutting Edge Applications and Scalable Production,” inMiami, Fla., April 10-11. The meeting emphasized advan-ces in commercial applications and the prospects for scala-ble production of carbon nanotubes. Srivastava also gave aninvited talk on plasticity and hydrogen diffusion on carbonand boron nitrate nanotubes, and Ames’ experimental nan-

otechnology researcher Allen Cassel spoke about nanotubeapplications in nanoelectronic circuitry and combinatorialchemical approaches for tuning the production of nano-tubes on different substrates. On his return trip, Srivastavaalso gave an invited talk, “Nanotechnology of MolecularMaterials, Electronics and Machines: Carbon Nanotubes,”at the Quantum Theory Project (QTP) seminar, at his almamater, University of Florida, Gainesville, on April 12.

Chimera Grid Tools Version 1.3Now Available

The Chimera Grid Tools (CGT) version 1.3, a packageof programs and scripts used to generate overset grid

Gridpoints 1www.nas.nasa.gov/gridpoints

Many years after winning the Nobel Prize for co-inventing the transistor, Walter Brattain said,“The only regret I have about the transistor is

its use in rock and roll.” On page 10, NAS researcherToshishige Yamada details his patent-pending research intoatomic chain electronics – a new, nanotechnology-basedapproach to device miniaturization that will revolutionizethe future of electronics. Reflecting on Brattain’s comments,I wonder what Yamada will say about his invention 30years from now?

Development of NASA’s prototype In-formation Power Grid (IPG) is progress-ing steadily. In April, the IPG’s firstimportant milestone to demonstrate datatransfer rates was achieved using datamining software. To meet the milestone,the IPG team retrieved data from fourremotely connected sites simultaneously.The success of our data mining demon-stration will allow the NAS team tomove forward in expanding the IPG’s

infrastructure, some of which will be demonstrated thisSeptember. The IPG project is directly related to a part ofthe NAS Division’s mission statement: “implementing andintegrating new high performance computing technologiesinto useful production systems.”

The NAS Facility is home to Lomax, a 512-processor SGIOrigin 2800, currently the world’s largest single-image sys-tem computer. Lomax supports the IPG and a variety ofNASA research efforts. NAS scientists are using the soft-ware program CART3D on Lomax to develop “virtualflight” through computational fluid dynamics. CART3Dgenerates grids to calculate flow fields around an air or

space vehicle. Recently, NAS researchers studied fourSpace Shuttle nose configurations, and the results will be“flown” in Ames Research Center’s Vertical Motion Simula-tor this fall.

The NAS Systems Division supports numerous NASAenterprises with high performance computing resources,and conducts research that will have far reaching impactson the NASA mission, the scientific community, and even-tually, on the public. For example, scientists from theNational Renewable Energy Laboratory are working to im-prove wind turbine designsusing wind tunnel testing andNAS computational modelingresources. Within the next 20years, this clean source of powercould provide more than 10 per-cent of the world’s electricity.

In addition, the NASA Re-search and Education Network(NREN) is improving theInternet with the global adop-tion of the Multicast Protocol.Multicasting will allow a singledata packet to be replicated anddelivered to multiple userssimultaneously, rather thaneach packet being deliveredserially. Through improved Internet speed, we will all ben-efit from multicasting’s applications.

As always, I welcome your feedback.Bill Feiereisenwfeiereisen@mail.arc.nasa.gov

From The Division Chief

NAS MissionTo lead the country in the re-search and development ofhigh performance computingfor NASA Programs andMissions by being the first todevelop, implement, and inte-grate new high performancecomputing technologies intouseful production systems.

To provide NASA and its cus-tomers with the most power-ful, reliable, and usable pro-duction high performancecomputing systems in thecountry.

NAS Researchers Lecture onCarbon Nanotubes

News From NAS

gp

systems for complex viscous computational fluid dynamics(CFD) problems, has been released. CGT’s new tools in-clude: surface and volume grid generation programs; gridmanipulation, smoothing, and projection programs; a mod-ular, general-purpose scripting system which can help build anentire grid system and produce Overflow flow solver inputs;and a graphical user interface known as Overgrid – usedfor generating grid systems. This version will significantlyreduce CFD cycle times for a wide range of applications.

Version 1.3 is the final release of the software that has beenfunded by the DoD Common High Performance Compu-ting Software Initiative CFD-4 project. Work is continuingtoward a fully automated grid generation process with theincorporation of an AI-based surface and volume grid gen-eration program, as well as the new, automated version ofthe overset connectivity software, Pegasus 5. CGT’s authorsare William Chan (NASA Ames), Stuart Rogers (NAS),Steve Nash (NAS), and Pieter Buning (NASA Langley).

News From NAS

INthe ongoing quest to solve challenges with creatingnanoscale molecular electronic devices, NAS re-

searchers are working to design nanoscale DNA diodes,transistors, and simple logic gates. The team is using aunique approach that provides a mechanism for self-assem-bly in future molecular-scale electronics, based on state-of-the-art nanotechnology and molecular electronics princi-ples. The research, conducted by Jie Han in the NASSystems Division, is funded by a two year, $80,000 grantfrom the Ames Director’s Discretionary Fund.

Much of the recent progress in computer technology isbased on creating smaller and smaller transistors for logicdevices and integrated circuits. Rapid advances have beenmade in fabricating and demonstrating individual electron-ic wires and devices. However, these new technologies stillface problems, including the requirement for devices toself-assemble.

Recent studies have shown that DNA has all the compo-nents needed to build an electronic device as well as theself-assembly characteristics needed to form complex elec-tronic circuits. Even more exciting to the research team,DNA has the unique feature of self-replication. The workwill focus on theoretical and computational proofs-of-con-cept, and experimental demonstration will be consideredusing nanotechnology and biology facilities at Ames.

Han is collaborating with the nanotube biosensor experimen-tal team at Ames and Nanogen, Inc., based in San Diego.

A scientific code used to simulate protein folding sequen-ces now performs 36 times faster than was previously

possible, thanks to code optimizations made by NASSystems Division researcher James Taft. The code, called

Ames Researchers DesigningDNA Electronic Devices

SGI Special: Lights, Camera, Action!An SGI film crew producing a corporate documentaryon the company’s supercomputers and their uses employ-ed the NAS Systems Division as a backdrop. AmesCenter Director Henry McDonald and NAS DivisionChief Bill Feiereisen were featured guests in the pro-duction. Feiereisen, above left in the visualization lab,discussed science conducted on SGI machines at NAS.Taping on June 13 lasted the entire day.

(Michael Boswell photo)

2 Gridpoints

gp

Corrections: In the Spring 2000 issue onpage 10, the caption for Figure 3 shouldhave read: “This image shows the sponta-neous attachment of a protein to themodified messenger RNA that encodes it.This allows an evolutionary approach tothe discovery of new functional proteinsin the laboratory. Researchers at HarvardMedical School start by generating tril-lions of random proteins in a single tube.Then, they select rare proteins exhibitinga desired function. This process is a mas-sively parallel analog computation of aproblem not currently solvable by serial digital computational

approaches. Current studies are directedtoward the discovery of functional proteinsin synthetically generated libraries of ran-dom-sequence proteins, rather than li-braries isolated from biological systems. Itis presumed that the proteins first utilizedby early living organisms were recruitedfrom such random-sequence libraries, andthat this study will indicate what propor-tion of such proteins are functional.(Credit: Glen Cho, Anthony D. Keefe,Rihe Liu, David S. Wilson, and Jack W.Szostak) Additionally, Ling-Jen Chiang’s

credit was omitted from the graphic on page 21.

Protein-folding Code HumsWith NAS Optimization

gp

COSMOS, is a molecular dynamics model used in NASA’sastrobiology research.

Taft modified the code to utilize 343 processors on Lomax,a parallel processor SGI Origin 2800 located at the NASFacility. With 512 CPUs, Lomax is currently the largest sin-gle-image system in the world. Using a shared-memory par-allel programming approach, Taft and COSMOS developerAndrew Pohorille, acting director of the NASA Computa-tional Astrobiology Institute, have been rebuilding the codeto run 30,000-atom problems on RISC microprocessor-based systems. Historically, many important moleculardynamics problems of this size have not scaled well on“clustered” parallel systems. Lomax, with very fast sharedmemory, is ideal for such computations.

“The results of this work can benefit the molecular model-ing industry in general because the newly developed numeri-cal techniques are applicable to many popular industrystandard models used by pharmaceutical research compa-nies in the U.S.,” says Taft.

The two-month optimization effort focused on insertingthe Ames-developed shared memory Multilevel Protocol(MLP) parallelization technique into the code. The MLPtechnique allowed dramatic improvements on the code’stwo most time-consuming routines. The overall speeduparises from scaling efficiencies in the MLP-based parallelalgorithm, coupled with greater reuse of encached data.

Members of the Ames Integrated Product Team onDevices and Nanotechnology – representing the

largest nanotechnology research efforts within the federalgovernment – received a group achievement award for“excellence in research and outstanding contributions to theemerging field of nanotechnology” at the 2000 NASAHonor Awards ceremony on June 7.

Project manager Meyya Meyyappan, credits the NASSystems Division as “the birthplace” of this work, originat-ing with a small theory group doing computational nan-otechnology at the NAS Facility. The team has expanded toinclude experimental research in the laboratory. Experi-ments on growing nanotubes for applications such as cancerdiagnostics attract a steady stream of visitors from the high-tech arena, including former House science and technologycommittee chair Bob Walker, shown at right, who touredthe “nanotube garden” on June 6.

Among those receiving the group achievement award wereNAS researchers M.P. Anantrim, Bryan Biegel, T.R. Govin-dan, Jie Han, Cun-Zheng Ning, Deepak Srivastava, AlexeiSvizhenko, and Toshishige Yamada.

www.nas.nasa.gov/gridpoints

Information & Technology ChairmanStephen Horn Tours NAS FacilityRep. Stephen Horn (R-CA, Long Beach), chairman of the106th Congress’ Subcommittee on Government Manage-ment, Information, and Tech-nology toured the NAS Facilityon April 24 before hosting ahearing at Ames on “EmergingTechnologies: Where is theFederal Government in theHigh Tech Curve?”

Division Chief Bill Feiereisenbriefed the congressman onNAS’s advances in high per-formance computing, nanotech-nology, and the InformationPower Grid. Horn then touredthe Visualization Lab and wasgiven a Virtual Mechano-systhesis demonstration byChris Henze and Bryan Greenof the NAS data analysis group.A carbon nanotube and a gra-phite T-junction were projectedon the Responsive Workbenchand Horn was able to interact with individual carbonatoms. At the conclusion of his tour, Rep. Horn said, “Thisamazing technology presents exciting possibilities for sci-ence and technology in the 21st century.”

— Nicholas A. Veronico

Senior research scientistChris Henze (right) briefsRep. Stephen Horn on theuse of NAS’s ResponsiveWorkbench in the NAS Vis-ualization Lab.

(Nicholas A. Veronico)

Gridpoints 3

Former Pennsylvania Congressman Robert Walker, center,and Ames Deputy Director of Information Systems PaulKutler, left, toured the Visualization Lab and were briefedon the capabilities of the 512 processor SGI Origin 2800,Lomax, by Bill Thigpen, NAS’s Engineering Branch chief.Walker, former chair of the House Science Committee,was on a fact finding mission at Ames on June 6. The triostand inside Lomax’s processor banks in the computerroom at NAS. — Nicholas A. Veronico

Fact Finding Tour of NAS And Ames

NASÕs Nanotechnology TeamMembers Recognized

gp

gp

4 Gridpoints

R esearchers Donovan Mathias and Jeff Onufer ofNAS’s computational physics and simulationgroup, tested four different nose designs for the

Space Shuttle on the 512-processor SGI Origin 2800,Lomax, during the weekend of March 25-26. “The goalof this project was not to find the best-suited nose designfor the orbiter, but to use computational fluid dynamicsto support virtual flight. We are developing a process tobring virtual flight into the early design phase, so the vehi-cle can be ‘flown’ before a single piece of metal is cut,” saysMathias who believes testing a design early in its develop-ment will save both time and money.

For the team’s work on Lomax,they employed a beta version ofthe CART3D (Three-DimensionalCartesian Simulation System forComplex Geometries) softwarepackage to generate the gridsnecessary for calculating flowfields surrounding the orbiter.Results from the Lomax tests willbe used in the Vertical MotionSimulator (VMS, see sidebar,NASA’s Full Motion Simulator,page 6) at NASA Ames in mid-September. The VMS will “virtu-ally” fly the Space Shuttle, evalu-ating flight performance of thenew orbiter designs.

During 21 hours of computationson Lomax, the team tested fourconfigurations – three new re-designs (created by Dave Kinneyfrom the Systems Analysis Branchat NASA Ames) plus the Shuttle’soriginal nose configuration. “Wewere able to study quite a num-

ber of the orbiter’s components during our weekend onLomax,” says Onufer. Each orbiter redesign took 60 simula-tions, requiring eight processors per simulation to study thecontrol surfaces, including the elevons (airfoils used tocontrol the movement of the shuttle), rudder, and speedbrakes. The team also examined flight conditions of thevehicle: Mach number, angle of attack, and angle of side-slip. This research was funded by the Information Tech-nology Program.

Once the orbiter grids were generated, they were uploadedto Lomax and used to calculate flow solutions using “Tiger,”

CART3D: A NewGeneration of Grids

3-D grid software automatically generates vehiclesimulations from complex geometries.

Mesh partitions flow over a Space Shuttle orbiter, with each color representing a set ofequations assigned to different CPUs. Sixteen CPUs were used for this calculation; thusthere are 16 divisions. Partitioning is done automatically on any number of processors.

(Aftosmis)

Gridpoints 5www.nas.nasa.gov/gridpoints

the flow solver module for CART3D. The Tiger code,developed by John Melton of Ames’ Aeronautical Projectsand Programs Office, models the inviscid flow fieldsurrounding a vehicle such as the shuttle orbiter.

Enabling Virtual FlightAiming to test the shuttle designs in the VMS, researcherswill merge their flow solutions calculated on Lomax withVMS’s existing mathematical model, which uses a set ofequations to describe the vehicle’s performance. Merging

the flow data with the simulation model, a pilot can evalu-ate the performance of the shuttle orbiter redesigns withhigh-fidelity simulations. “Ideally, we would like real astro-nauts to fly the simulation to provide the most useful feed-back on the flight performance of the vehicle,” says Mathias. Currently, NAS research scientist Michael Aftosmis is work-ing on a scalable flow solver to replace the Tiger code.Aftosmis’ anticipated flow solver will drastically reduce thetime necessary to compute flow fields by distributing themesh generated onto different CPUs. Mesh decomposition

Grid Generation Using CART3DNAS research scientist Michael Aftosmis, John Meltonfrom Ames’ aeronautical projects and programs office,and Marsha Berger, co-director of the Courant Instituteat New York University, released CART3D in 1997,after five years of development. CART3D is a preliminarydesign analysis package created to solve the Euler Equa-tions of Motion with Cartesian grids around complexvehicles. Aftosmis’ work is funded by NASA’s HighPerformance Computing and Communications(HPCC) Program.

CART3D can import geometry from many differentsources, including most major computer-aided design(CAD) packages, and com-putes a complete flow sim-ulation around any com-plex vehicle. The softwaretakes the surface definitionof a vehicle and generates amesh, or set of grids, neces-sary to compute the airflow around it using a flowsolver. “The way CART3Dsees geometry is unique –parts of a whole can bemoved, and the entire con-figuration does not have tobe regenerated. The soft-ware uses a component-based approach enablingthe automatic readjustmentof the grids, saving a ton oftime,” Aftosmis says.

The efficiency of CART3D has increased since it wasfirst released. “A majority of a researcher’s time wasbeing consumed in converting CAD geometry intosomething that could be used for mesh generation.CART3D has reduced the execution time of this step,and the ultimate goal is to put the whole thing in a box,eliminating the human element,” says Aftosmis.“CART3D is a very robust software package, and can be

quickly applied to almost any air or space vehicle,” addsresearcher Donovan Mathias. “Grids are essentially gen-erated in a matter of minutes on just about any UNIXor PC platform using CART3D.”

The sequence of grids (A through D, below), surroundstwo advanced destroyer-type surface ships. These simula-tions were generated using CART3D’s flow solver mod-ule, which employs a multigrid technique to improve theefficiency of the simulation. This method utilizes asequence of coarse meshes to reduce the calculation timeof the final solution of a detailed, highly refined mesh.The amount of computational work performed is pro-

portional to the complexity of the mesh – less complexgrids enable rapid computations using solution estimates.This multigrid technique uses a hierarchical approach –the coarse solution estimates, (figures A through C), areused to “drive” the complete solution on the fine mesh(Figure D). The grids are automatically produced anddistributed over Lomax's processors. In this example, thefinest mesh has 4.5 million grid points while the coars-est has only 4,500. (Aftosmis and Mathias)

A B

DC

6 Gridpoints

enhances the solver’s scalability on parallelcomputers like Lomax by explicitly assign-ing different parts of the problem to themachine’s processors. An application usingthe new solver with the same number ofCPUs will work five to 30 times fasterthan the “Tiger” code. “We are reallylooking forward to putting this newsolver to work – it will be a great dealfaster than Tiger,” says Onufer. Aftosmisplans to release the new code to a selectgroup of government users in mid-Juneand apply it to high-speed applications:“We want to stretch the limits of the newsolver by applying it to complex problemsin all three flight regimes; subsonic, tran-sonic, and supersonic.”

CART3D’s FutureMore than 80 institutions, including theU.S. government, industry, and academiacurrently use CART3D. Mathias, Onufer,and Aftosmis plan to continue their workon improving flow calculations for flightsimulations in the future. “I think virtualflight is a great application of simulationsoftware. The application stresses CART3Dbecause it requires a lot of throughput,but is very doable with machines likeLomax,” says Aftosmis.

—Holly A. Amundson

Above: The velocity component of the baseline shuttle orbiter as computed onLomax. The angle of side-slip is zero degrees, and angle of attack is 12 degrees.Pink areas represent free stream air at MACH 0.9, while the darker colors corre-spond to a lower air-stream velocity. The blue area behind the shuttle representslow velocities. (Onufer, Mathias)

Right: The pressure coefficient used to com-pute lift and drag is depicted on the baselineshuttle orbiter. Red represents high pressureareas, while the darker colors denote lowpressure. (Onufer, Mathias)

gp

Space Shuttle configurations calculated on the NASFacility’s Lomax high performance computer willtake flight in Ames Research Center’s Vertical

Motion Simulator in September. Each of the four newlycomputed Space Shuttle designs will be flown in the simu-lator, which is capable of replicating vertical, lateral, andlongitudinal movements as well as roll, pitch, and yaw.

The Vertical Motion Simulator’s size and range of mo-tion are key to producing high fidelity flight simulations.Housed at the Vertical Motion Flight SimulationLaboratory in a ten-story building, the simulator’s verti-cal platform is 70 feet wide and can be elevated nearly 75feet. On top of the vertical platform sits the inter-changeable simulator cab where pilots can explore acraft’s flight envelope.

The lab houses five different cabs that can be built upinto a wide variety of helicopter, single-seatfighter jet, tilt rotor and transportaircraft, and the Space ShuttleOrbiter. Each cab can beconfigured to the researchneeds of various test projectsby changing the seats, flightcontrols, and cockpit instru-mentation. An image presen-tation system gives the pilota virtual out-the-windowview of the world.

For further information on the AmesAviation Systems Division’s VerticalMotion Flight Simulation Laboratory, visit:http://www.simlabs.arc.nasa.gov/vms/vms.html.

NASA’s Full Motion Simulator

VMS/

Nees

ham

Wal

lace

, Je

nnife

r Go

udey

Gridpoints 7www.nas.nasa.gov/gridpoints

Since 1998, the NASA Research and Education Net-work (NREN) – a component of the NAS SystemsDivision, has been a leader in the global deployment

of new multicast routing and forwarding protocols used forsending information across the Internet. Protocol Indepen-dent Multicast–Sparse Mode (PIM-SM), Multicast SourceDiscovery Protocol (MSDP), and Multicast Border Gate-way Protocol (MBGP) are now being adopted worldwide asthe standard protocols for multicast data trans-fer. Multicast will improve the Internet’s effi-ciency, and expand networking technology toallow scientists and researchers to collaboratewhile conducting experiments from remotelocations. In the future, commercialization ofmulticast technology will benefit society as awhole, enabling advances in aerospace research,astrobiology, and other Earth sciences, as wellas telemedicine.

How Does Multicasting Work?Within the Internet, packets (pieces of data) areusually sent one at a time. This method, called“unicasting,” is a simple mechanism for one-to-one communication. However, for one-to-many communication, unicasting can createextreme network congestion. With multicast,packets are simultaneously transmitted to mul-tiple sites at the same time. As a result, multi-casting conserves bandwidth, reducing theinherent bottlenecks created by one-to-manytransmissions. During multicast, only one copy

of the data is transmitted while network routers replicatethe packets when needed. As a result, on both the sendingand receiving side of the transmission, network bandwidthis conserved.

Multicasting is similar to broadcasting radio signals; like aradio signal, multicast packets originate from a single site.Anyone with the correct equipment can receive the signal,

NASA Research andEducation Network Drives

Global Deployment ofMulticast Protocol

Multicast will speed up the Internet by enabling a single datapacket to be distributed to multiple Internet users simultaneously.

By Hugh LaMaster and Kenneth Freeman

Source

H-3

Router Router

Network A Network B

10 Mbps

Router

H-2

H-1

Router Router

Router

Figure 1. If the source computer is multicasting data at 10 Megabits persecond (Mbps), 10 Mbps will be transmitted between Network A and NetworkB, reaching Host H-1 and Host H-2 simultaneously. Without multicast, 20Mbps of data would be required to transmit data to Host H-1 and Host H-2,doubling the transmitted data time. (LaMaster and Freeman)

8 Gridpoints

while radio signals pass by any-one who does not want to re-ceive the signal, or does not havethe right equipment. Similarly,multicast is “broadcast” to thenetwork but is only received bythose interested in the data.

Evolution of Global IP MulticastWith the explosive growth of theInternet, networks are strugglingto keep up with the increasingdemand of millions of new users.In fact, the World Wide Web isoften referred to as the “WorldWide Wait.” To reduce the amount of repetitive data trans-missions, global IP (Internet Protocol) multicast technologyallows a single packet stream to be distributed across theInternet simultaneously to any number of recipients. Multi-cast is often the only way to simultaneously distribute real-time data to numerous recipients efficiently. It also allows adata stream to be replicated at intermediate routers so thatany link only receives a single copy, regardless of how manydownstream recipients have requested the information.

Ames Research Center has been a focal point of IP multi-cast deployment since the early days of MBone, the multi-cast backbone. Global IP multicast routing, originallydone via the distance vector multicast routing protocol(DVMRP), was the basis for the MBone. One of the firstsuccesses in using multicast was NASA’s Space Bridge toRussia in 1996, where multiple remote sites were fed infor-mation simultaneously.

More recently, NREN has deployed a number of new net-working protocols. Collaborating with the Ames GatewayFacility (Ames Research Center’s data, video, and telecom-munications relay point), Cisco Systems, and Sprint’s re-search network (Sprint-ICM), the first Multicast InterneteXchange (MIX) was created; the first multi-networkdeployment of the Multicast Border Gateway Protocol(MBGP – a routing protocol that replaces DVMRP forinter-domain routing) was achieved; and the first multi-domain global deployment of the Protocol IndependentMulticast (PIM – a forwarding protocol) was accomplished.

In July 1998, NREN began deploying the Multicast BorderGateway Protocol (MBGP). Initial sites included NREN,Sprint-ICM, and a transit-AS (Autonomous System). By

fall 1998, connections to the Department of Energy’s Ener-gy Sciences Network (ESNET), the Defense Research andEngineering Network (DREN), as well as Internet serviceproviders Verio, Exodus, and AboveNet, had been added.During this time, NREN began “peering” with vBNS(National Science Foundation and MCI’s very highspeedBackbone Network Service) in Chicago. When Internet2’sAbilene Research Network joined, all of the federal gov-ernment’s Next Generation Internet networks hadadopted MBGP.

Multicast Source Discovery Protocol DeploymentThroughout the Mbone and early MBGP deployment,dense-mode protocols were used that can sometimes floodunnecessary data across wide area network (WAN) links.Using dense-mode, there was no way that very high band-width sources exceeding 10 Megabits per second (Mbps)could be supported across typical WAN links.

Beginning in the fall of 1998, NREN initiated testing of anew protocol, Multicast Source Discovery Protocol(MSDP), allowing the use of sparse-mode protocols global-ly. Sparse-mode enables multicast to be used without specif-ic bandwidth limitations, and supports very high band-width data streams.

In spring 1999, driven by the requirement to support theVirtual Collaborative Clinic – an Internet-based telemedi-cine demonstration – NREN coordinated the deploymentof MSDP and PIM-SM with the Internet2/Abilene net-work, the Corporation for Education Network Initiativesin California’s CalREN-2 network, as well as StanfordUniversity and the University of California at Santa Cruz.The deployment of PIM-SM and MSDP enabled data

NRENPIM-SM

AboveNetAT&TAxientExodusFrontier Global CenterLevel3Lightning / GigabellQwestTeleglobeVerio

AboveNetDigexFrontier GlobalcenterIDTQwestVerio

vBNSANLMerit

NCSANORDUnet

SURFnetTransPac

ESnetDREN

Sprint - ICM

Ames MIX

MAE-West

Mbone

UUnet

NISN

FIX - West

Abilene

NGIX - WestAmes

Jet Propulsion Lab

ChicagoGlenn

Goddard

Langley

Marshall(Future Multicast Site)

Figure 2. As of June 2000, NREN’smulticast architecture has beenadopted by numerous networksacross the nation. Marshall SpaceFlight Center has been designatedas a future multicast site.

Gridpoints 9www.nas.nasa.gov/gridpoints

transfer rates of at least 24 Mbps to all sites (50 Mbps tosome sites).

During summer 1999, the deployment of MSDP andPIM-SM continued throughout Internet2’s customernetworks, and among commercial Internet serviceproviders. NREN provided the catalyst for this deploy-ment by using PIM-SM and MSDP at the Ames MIX,as well as its peerings at Chicago. By January 2000, theconversion from dense-mode to MSDP/ PIM-SM wasalmost complete throughout the United States (seeFigure 2, opposite).

Multicast Routing ProtocolsGlobal multicast routing requires routers to replicatemulticast packets and forward copies to each interfacewhere a downstream receiver has requested a copy. Thisprocess is known as multicast forwarding, and itrequires the creation of a multicast forwarding tree foreach multicast group.

In addition to the creation of a forwarding tree, a rout-ing protocol is required. In the usual configuration, theAS (Autonomous Systems) Interior Gateway Protocol(IGP) is the source of unicast routes that can also be used formulticast. A common IGP is OSPF (Open Shortest PathFirst). Among autonomous systems, an Exterior GatewayProtocol (EGP) is used to exchange routes. Today, the uni-versal EGP is Border Gateway Protocol-4 (BGP-4). Multi-cast Border Gateway Protocol (MBGP) is an extension toBGP-4 that allows for an exchange of additional routesspecifically for multicast. This is required because the datapaths for unicast and multicast can differ.

In order for multiple autonomous systems to know abouteach other’s multicast sources, a helper protocol is required.Today, this protocol is Multicast Source Discovery Pro-tocol (MSDP). MSDP floods information about multicastsources among all the Rendezvous Points (RPs) in multi-ple domains, allowing receivers in one autonomous systemto receive data from a different autonomous system (seeFigure 3).

Protocol Independent Multicast-Sparse ModeSparse protocols use the explicit join of a multicast groupby a host to trigger the creation of the multicast forwardingtree. Initially, the first-hop router joins the shared-tree fromthe rendezvous point. For high-bandwidth sources, it willthen switch to a shortest-path tree from the first-hop routerback to the source. PIM-SM does this through the use ofRendezvous Points that coordinate the creation of the mul-ticast distribution trees throughout the PIM-SM domain(usually an AS). Within a defined network area, the RP cre-ates a shared tree for every multicast group, and it coordi-nates the creation of a shortest path tree from a source to allrecipients for high-bandwidth sources.

The Future of MulticastMSDP has always been regarded as an interim protocol forseveral reasons. It incorporates a feature for backwards-compatibility with the old Mbone, and it floods sourceinformation. This flooding, while vastly superior to thedense-mode data flooding it has replaced, still will not scaleto an Internet with very large numbers of active sources.For some time now, development of global shared-treeprotocols has been undertaken. Today this has to be consid-ered a difficult and as yet unsolved problem.

In addition, with the commercialization of the Internet, anew constituency of commercial multicast Internet broad-casters is appearing. These broadcasters would like tosupport broadcast-only functionality, rather than themany-to-many multicast model. They also would like asingle broadcaster to be able to deliver very large numbersof channels, which would create a multicast address alloca-tion problem. A segment of the multicast address space hasbeen allocated for this so-called source-specific multicast,which will allow both of these problems to be solved whilemaintaining most of the multicast address space for many-to-many communication. Source-specific multicast willrequire a simple modification of PIM-SM behavior toaccommodate the different functionality. MSDP will not berequired for source-specific multicast because the sourceswill typically be announced via the Web.

Today, IP Multicast is undergoing rapid development anddeployment in the commercial world, and will be a keytechnology in providing efficient sharing of real-time dataamong large numbers of systems.

Network B

Router Router

Network A

Router

Router

RendezvousPoints

Network C

Router

RouterRendezvous

Points

RendezvousPoints

Figure 3. With inter-domain multicast routing, all routers withexternal MBGP connectivity have parallel MSDP connectivity. Allrouters with external MBGP and MSDP connectivity have meshedinternal MBGP and MSDP connectivity. In addition, the RP must bepart of the MSDP mesh. Each network represents an AutonomousSystem (AS). Moreover, the red arrows depict links with MBGP andMSDP connectivity.

gp

10 Gridpoints

Since its invention in 1948, the transistor has revolu-tionized everyday life – transistor radios and TVsappeared in the early 1960s, personal computers came

into widespread use in the mid-1980s, and cellular phones,laptops, and palm-sized organizers dominated the 1990s.The electronics revolution is based upon transistor minia-turization; smaller transistors are faster, and denser circuitryhas more functionality. Transistors in current generationchips are 0.18 µm (micron) or 180 nanometers in size, andthe electronics industry has completed development of0.13 µm transistors that will enter production within thenext few years. Industry researchers are now working to re-duce transistor size to 0.1 µm – a thousandth of the widthof a human hair. However, studies indicate that the minia-turization of silicon transistors will soon reach its limit.

For further progress in microelectronics, scientists haveturned to nanotechnology to advance the science. Ratherthan continuing to miniaturize transistors to a point wherethey become unreliable, nanotechnology offers the newapproach of building devices on the atomic scale (see Tran-sistors: Electronic Workhorses, opposite). One vision for thenext generation of miniature electrical devices is atomicchain electronics, where devices are composed of atomsaligned on top of a substrate surface in a regular pattern.The Atomic Chain Electronics Project (ACEP), part of thesemiconductor device modeling and nanotechnology group

integrated product team within the NAS Systems Division,has been developing the theory for understanding atomicchain devices. In this type of research, collaboration with

The SmallestNanoelectronics:Atomic Devices

with PreciseStructures

Transistor miniaturization isreaching its limit. Atomic scale

transistors are being developed toenable future microelectronics.

By Toshishige Yamada

Larger Than 1 Micrometer

0.01 to 0.1 Micrometer

Gate

Gate

Source Drain

Source Drain

1a

1b

1c Smaller Than 0.001 Micrometer (1 Nanometer)

Figure 1: The evolution of transistor miniaturization hastaken electronic devices from the macroscopic scale (1a),greater than 1 µm (micron). An electron particle (green)sees a jelly area (blue) formed by 10,000 dopant atoms,which is averaged out to be the same in devices, resulting inthe same switch-on voltage. In transistors miniaturized tothe mesoscopic scale (1b), 0.1 to 0.01 µm, an electron wave(green) can tell precisely where the dopant atoms are locat-ed, and this leads to different switch-on voltages. To solvethis problem, researchers at the NAS Facility have patentedan atomic device concept using atomic chains (1c) ratherthan further the miniaturization of conventional transistors.

Gridpoints 11www.nas.nasa.gov/gridpoints

the experimentalists is also essential for success. ACEP mem-bers are collaborating with Dongmin Chen of the Nano-physics Group at Boston’s Rowland Institute, who charac-terizes electronic properties of atomic structures using aScanning Tunneling Microscope (STM). ACEPresearchers are also collaborating with RichardA. Kiehl at the University of Minnesota’s elec-trical engineering department, who studies theautoformation of atomic structures. Auto-formation may lead to a breakthrough formass production of atomic-scale devices.

Doping and Gain in TransistorsA critical element in transistor functioning is the use ofdopants – impurities added to silicon to manipulate andcontrol its electronic properties. Typically, there are several

Millions of tiny transistors are at the heart oftoday’s electronics world. The main function ofa transistor is to amplify a small input signal

into a large output signal as shown in the diagram (a).The most common transistor used in today’s electronics isa field-effect transistor (FET). To understand the opera-tion of an FET, the popular analogy is that of a flexiblehose system with weight on it. When the weight is small,water flow is not hindered, as shown at right. As theweight is increased, the hose becomes pinched, resultingin less water flow. In this analogy, the weight correspondsto the input signal and the water passingthrough the hose is the output.

In a field-effect transistor, the input is a volt-age applied to the gate electrode, and theoutput is a current flowing between thesource and the drain. When the gate voltageis low, electrons flow freely from the sourceto the drain. When the gate voltage is high,electrons are repelled under the gate andfewer electrons flow. This is known as thefield effect.

In the analogy of water flow through a hose(b), the constriction caused by the weightdepends on the material of the hose. Forexample, if the hose is a steel pipe, then theweight (input) has little effect on flow (out-

put). In the same way, if the substrate of an FET is ametal, the gate voltage has little effect on current flow.Therefore, we need a material corresponding to a flexiblehose – a semiconductor.

What FET property corresponds to the flexibility of thehose? That is doping, or the intentional inclusion ofimpurity atoms in the semiconductor substrate. Bymanipulating doping, a semiconductor can be changedfrom something almost an insulator (softest) to some-thing almost a metal (hardest).

Figure 2: Rendering (2a) represents an atomi-cally precise structure using the substrateatomic lattice (blue atoms as a template foratomic chain electronics (red atoms). Two con-finement schemes for topmost atoms on thesubstrate are (2b), where atoms are floating,and (2c), where the atoms are firmly bonded tothe substrate. Green dots represent hydrogenatoms, and they work as an electronically insu-lating layer with an atomic thickness.

2c Chemical Bonding

Precise Atomic Structure

2b Floating

2a

Out

Gate

In

More Current

MoreWater

LessWater

Weight

Less Current

(b) Hose vs FET

(a) AmplificationInput

Output

Out

ChannelChannelSource DrainSource Drain

SubstrateSubstrate

TransisitorTransisitor

Transistors: Electronic Workhorses

12 Gridpoints

thousand dopant atoms in a transistor and these appear as adopant “jelly” as in Figure 1(a) – on page 10. When usinglarge numbers of dopant atoms in the construction of anelectronic device, the precise location of these atoms in thetransistor is not very important to the functioning of thetransistor. However, when the transistor size is reduced toless than 0.1 µm as in Figure 1(b), the number of dopantatoms is small – less than 100. At 0.1 µm or less, each do-pant’s position affects how electrical current flows throughthe transistor. Present-day manufacturing technology can-not precisely control the location of so few dopant atoms,and this causes small variations in how the transistor func-tions. This is fatal when millions or billions of transistors

are integrated on a computer chip,because these slight variations cancascade from one device to thenext and eventually lead to mal-function. The solution to thisvariation problem devised byACEP calls for all device struc-tures to be built using atomicchains laid out in a regular, precisepattern by anchoring atoms to asubstrate as in Figure 1(c).

A second critical element in tran-sistor function is that the output ofthe device should be a magnifica-tion of its input – this is calledgain. To create atomic chain tran-sistors that produce gain, ACEPresearchers have found that thephenomenon known as field effect(see Transistors: Electronic Work-horses, page 11) can be adopted as

an operating principle. To build a field-effect transistor atthe atomic scale (similar in construction to those of largerscales), scientists need to devise chains with both semicon-ductor and metal properties. First, the ACEP team tackleda simple problem theoretically: If researchers can arrangesilicon atoms along a line floating in air, is this chain semi-conducting? The answer is surprising: although bulk or thinfilm silicon is semiconducting, the silicon chains are alwaysmetallic. Fortunately, ACEP’s research shows that a magne-sium chain is semiconducting, even though ironically, bulkmagnesium is metallic.

Substrate Determines Chain PropertiesOf course, atoms cannot be floated in air, but must beplaced on top of a substrate. Surface atoms of the substrateattract atoms in the chain and hold them at fixed positions.The substrate has atomic scale corrugations on the surfaceand these may be used to create a precise pattern as shownin Figure 2(a) – on page 11. However, ACEP membershave found that this force is too weak to secure atoms reli-ably, see Figure 2(b). ACEP scientists concluded that achemical bonding scheme to secure atoms to the substrateas in Figure 2(c) was the reliable approach to follow. Re-search soon showed that the chain properties were stronglyinfluenced by the substrate material and surface orientation

Germanium topmost atom

Hydrogen atom

First Layer Silicon atom

Second Layer Silicon atom

Electron 'cloud'

π-chain

σ-chain

Figure 3. SemiconductingGermanium chains electronicallyisolated from a silicon substrate.

A

A4c

4b4a

Figure 4: Atom manipulation and electronic characterizationof small atomic structures are accomplished using a ScanningTunneling Microscope (STM) tip (green). Researchers canpush or pick up an atom (red) on the substrate (blue) usingan STM tip as in (a). A created small atomic structure (red)can be electronically characterized by measuring current forapplied voltage either in (b) a vertical geometry, or (c) ahorizontal geometry. The latter needs a dual-tip STM and istechnically quite challenging.

Gridpoints 13www.nas.nasa.gov/gridpoints

when chemical bonding is relied upon. In fact,the same atomic chain can be either metallicor semiconducting depending on the configu-ration of the substrate. For example, siliconchains with two chemical bonds to the sub-strate atoms are semiconducting – otherwisesilicon chains are metallic. Aluminum chainswith one chemical bond are semiconducting –otherwise, aluminum chains are metallic. A further concernwas that electrons travelling along an atomic chain maydetour into the substrate through chemical bonds and pos-sibly exit via a neighboring chain – resulting in the short-circuit of two atomic chains. This short circuit is fatal forelectronic applications and must be avoided. ACEP scien-tists have clarified the conditions for which the substratesurface acts as an insulator in the chemical bondingscheme. According to this, silicon and germanium crystalsoffer good insulating substrates.

Putting It All TogetherThe ACEP team has achieved the first step towards thedevelopment of atomic chain electronic devices. On top of

a silicon substrate, germanium atoms can be attached withtwo chemical bonds as shown in Figure 3 – on page 12.This structure is a semiconducting chain on an insulatingsubstrate. Nearby chemical bonds on the substrate are satu-rated with hydrogen atoms so that they are electronicallyinactive. This has resulted in the development of an essen-tial component for an FET on the atomic scale.

These structures could be experimentally fabricated andcharacterized using an atom manipulation technology witha Scanning Tunneling Microscope (STM) as in Figure 4 –on page 12, or some other related techniques. Quite recent-ly, the field effect in the molecular/atomic scale has beendemonstrated in carbon nanotube FET experiments inFigure 5 by Cees Dekker’s group at Delft, and PhaedonAvouris’ group at IBM. This supports the ACEP’s predic-tions for atomic devices, and the basis for atomic chainelectronics is now being established.

AcknowledgementsThe author acknowledges fruitful discussions with and/orcontinuous support for this project by Meyya Meyyappan,T. R. Govindan, C. W. Bauschlicher, Jr., H. Partridge, andB. A. Biegel. Careful reading by H. Yamada is gratefullyacknowledged.

Toshishige Yamada is a Senior Research Scientist withComputer Sciences Corp. at NASAAmes Research Center. He obtained aPh.D. in electrical engineering fromArizona State University. He heldpositions at NEC MicroelectronicsResearch Labs., Japan, and StanfordUniversity. NASA filed a patentapplication on his atomic chain elec-tronics innovation, and the work waspresented at an invited talk in the196th Electrochemical Society Meeting last fall.

Visit these sites for additional sources ofinformation about nanotechnology:

Integrated Product Team on Devices, Nanotechnologywww.ipt.arc.nasa.gov

The Integrated Product Team on Devices and Nanotech-nology’s home page offers a gallery of nanotechnologyexperiments, NASA Ames’ nanotechnology capabilities,and a list of projects and publications.

National Nanotechnology Initiative Home Pagewww.nano.gov

Read the full text National Nanotechnology Initiative:Leading to the Next Industrial Revolution, supplement toPresident Clinton’s FY 2001 budget.

Interagency Working Group on Nanoscience,Engineering and Technologywww.whitehouse.gov/WH/EOP/OSTP/NSTC/html/iwgn/IWGN.Research.Directions/toc.htmlThis document lays the foundation for the future ofnanotechnology research and development, and howresearch efforts should be coordinated within the scientific community.

_

+

NanotubeID

VG

_

+ VD

Source DrainSource Drain

Gate (silicon)Gate (silicon)

SiO2SiO2

Figure 5

gp

Figure 5: Carbon nanotube field effect transistors(FETs) experimentally built by Delft and IBM, withfunction similar to that predicted for an atomicchain FET in Fig. 1(c), have demonstrated thetransistor action in the molecular/atomic dimen-sion successfully.

14 Gridpoints

Wind-generated electricity is the fastest growing,most cost-competitive renewable energy resourcein the world. According to Wind Force 10, an

international report by the European Wind Energy Associ-ation, by the year 2020, wind-generated electrical powerwill provide 10 percent of theworld’s electricity, reducingcarbon dioxide emissions bymore than 10 billion tons ayear. Unlike fossil fuels,renewable energy does not ex-haust the Earth’s supply ofnon-renewable mineral re-sources such as coal and natu-ral gas. A team of researchersfrom the Department ofEnergy’s National RenewableEnergy Laboratory (NREL),NASA Ames Research Center,and the University of Califor-nia at Davis have launched aresearch program using bothexperimental and compu-tational approaches to improveour understanding of windenergy systems. Researchersfrom Ames’ Army/NASARotorcraft Division are usingthe high-performance comput-ers at the NAS Facility to sup-port the project.

The experimental part of thisprogram, known as theUnsteady Aerodynamics

Experiment (UAE), tested NREL’s wind turbine in theNASA Ames 80- by 120-foot wind tunnel from mid-Aprilto early May. Computational fluid dynamics (CFD) pre-dictions for the same wind conditions gathered during thewind tunnel experiment were computed on the CRAY

C90 operated within the NASDivision and managed by theConsolidated SupercomputingManagement Office (CoSMO).These computations require largeamounts of data storage availablethrough NAS’s SGI Origin-basedstorage system, NAStore.

Computational ModelingNREL’s wind turbine experimentis the first in the United States toemploy high-resolution Navier-Stokes calculations to investigatethe aerodynamics of a wind tur-bine. Using CFD predictions,NREL and its collaborators planto develop blade stall delay mod-els for creating better turbine de-signs applicable to both the windpower and rotorcraft industries.

“Some problems encounteredwhen designing an efficient windturbine are related to those expe-rienced with rotorcraft. Bothwind turbines and rotorcrafthave rotating blades with similarstall behavior,” explains BobKufeld, NASA project lead for

CFD Supports WindTurbine Research

Before any testing took place in the wind tunnel atNASA Ames, experiments were conducted at NREL’sfield labs in Boulder, Colo. Researchers installedsmoke generators in the tip of the wind turbine rotorto experimentally visualize the airflow and turbu-lence surrounding the blade. (NREL)

NAS provides the National Renewable EnergyLaboratory with computational resources to help

improve next-generation wind turbine designs.

Gridpoints 15www.nas.nasa.gov/gridpoints

the UAE’s wind tunnel experiments. Stall occurs when therotor lift no longer increases with the angle of attack.Using results obtained from the well-known CFD codeOVERFLOW, along with wind tunnel data, researcherscan gain a better understanding of the rotor aerodynamics.

To improve wind turbine design, rotor design codes usedby industry need a high degree of accuracy at low compu-tational cost. “To date, there hasn’t been a thorough set ofexperimental data to verify the accuracy of the computercodes because wind turbines operate in the inherently tur-bulent wind of the Earth’s atmosphere. Current engineer-ing models used are not very effective at higher wind speedsbecause our present theories do not accurately predictblade stall,” explains Earl Duque, computational lead forthe UAE in the Army/NASA Rotorcraft Division. Resultsfrom the UAE were obtained in the precisely controlledwind tunnel environment, and will provide industry with avaluable database to validate its computer models andCFD codes. This new data will enable designers to developmore reliable and cost-effective wind turbines.

Duque’s extensive background in rotorcraft computationsis beneficial for this project. “My experience working withothers on rotorcraft computations in the past has madethis project easier in the areas of visualization and grids.”With the help of MIT researcher David Kenwright, for-merly of NAS’s data analysis group, Duque modeled vortexpatterns over a rotorcraft blade, using Kenwright’s flowvisualization techniques. This work helped him visualizewhat is happening with the wind turbine blades. Workingwith NAS senior scientist Rupak Biswas, Duque alsoinvestigated grid adaptations, which enables accurateprediction of blade stall. Duque’s earlier studies allowedhim to understand how to build the grid systems for theNREL experiment.

Reducing the Cost of EnergyCharged by the Department of Energy to reduce the costof electrical power produced from renewable sources, NRELbegan its wind experiment in 1989. The UAE researchwind turbine has been field-tested in various configurationssince the beginning of the project. “The current phase ofthe experiment (phase six) is the most realistic wind tur-bine blade – it’s a two-blade rotor system, which is moremodern and cost-effective than the three-blade configura-tion of earlier experiments. Two blades require less build-ing materials but generate the same amount of power,”says Kufeld.

When NREL began its studies, wind energy cost morethan 20 cents per kilowatt-hour – which includes the sys-tem’s initial capital investment, operation, and mainten-ance costs. Today, the price has dropped substantially toabout four cents per kilowatt-hour. NREL’s goal is toreduce this cost further, to two and a half cents per kilowatt-hour.

Comparisons: Actual vs. ComputationalThe wind tunnel test turbine is equipped with cameras, sur-face pressure taps, and strain gauges to gather data for vali-dating numerical prediction models of aerodynamic bladeresponses. Testing in a controlled wind tunnel environmentprovides data that is free of interference from atmosphericturbulence, which substantially complicates turbine bladeaerodynamic response. This data is ideal for validating com-putational blade stall models and CFD codes.

Continued on page 16

Representing the air surrounding the wind turbine blade, thesevisualizations of numerically simulated flow fields are based onwind tunnel test conditions of NREL’s phase-six wind turbine.The visualizations were constructed using Fieldview Software, byIntelligent Light Inc., on a desktop computer. The red bandsrepresent streamlines flowing over the blade at a single instantin time. Before the blade begins to stall (above), the airstreams smoothly over the blade’s surface. When the bladebegins to stall, the airflow becomes unsteady, limiting poweroutput (below). (courtesy Earl Duque)

Stalled and Non-Stalled Turbine Blades (stream lines)

Continued from page 15Working closely with NREL, Duque broughtNAS computational resources to the project in1997. “Computations extend the boundariesof what we can test,” says Duque. “In the fieldor the wind tunnel, there are only so manythings you can measure. With CFD you cansee what’s happening to the airflow around theblade that cannot be seen in field experiments.”Only a few selected points on the rotor havesensors to collect data in an experiment. If thedata from those points match well to comput-er calculations, every point of the flow fieldcan be re-created using CFD.

Jasim Ahmad and Roger Strawn in the Army/NASA Rotorcraft Division have improved thealgorithm that Duque used to run CFD calcu-lations of the wind turbine in phase two of theexperiment. Phase two computations took 150CPU hours to complete. Now, phase six calcu-lations can be done in 40 CPU hours onCoSMO’s CRAY C90.

Duque will use the experimental data collectedfrom the wind tunnel test and compare it tothe predictions he creates using OVERFLOW(version 1.8), a Reynolds-averaged thin-layerNavier-Stokes (RaNS) code. He plans to useOVERFLOW-D for the next set of wind tur-bine computations. The first version of OVER-FLOW, developed by Pieter Buning, currentlyat NASA Langley Research Center, was latermodified by Ahmad for unsteady and steadyrotor applications. OVERFLOW predicts thepower output and the flow field around a tur-bine. Strawn is consulting with NAS research-ers Rupak Biswas, Rob Van der Wijngaart, andMaurice Yarrow on a parallel version ofOVERFLOW-D for rotorcraft.

In the future, OVERFLOW-D may be usedby the Information Power Grid (IPG) com-munity (see the IPG story on page 18) forother fluid flow problems requiring highlyrefined grids, resulting in massive amounts ofcomputational data such as helicopter and tilt-rotor configurations.

“OVERFLOW really tests the IPG system –there is too much data to process on any singlemachine at NAS, but it can potentially bedone on the IPG once it is fully operational,”says Strawn. The team anticipates runningOVERFLOW on an SGI Origin 2000 for thephase six calculations in winter 2001.

The graphic above represents the velocity contours of an S809 airfoilunder pre-stall conditions. This airfoil is the same one used on NREL'sUnsteady Aerodynamics Experiment wind turbine. Due to its design, theS809 airfoil was calculated to begin stalling at approximately 10-degreesangle of attack and to hold lift over a large range of angles. This liftcharacteristic helps maximize and control the power output of the tur-bine. Below, the airfoil is positioned at a 14-degree angle of attack, apost-stall condition. Wind turbines use stall to control power output of asystem. Unit of measurement is Mach number. (Earl Duque)

Wind Turbine Airfoil(pre- and post-stall conditions)

16 Gridpoints

Gridpoints 17www.nas.nasa.gov/gridpoints

NREL, NASA Ames, and UC Davis have made measura-ble progress in the area of wind power production by pro-viding the industry with detailed experimental data andvalid computational methods. From gathering detailedpressure data on wind turbine blades to predicting thesame pressure on CoSMO’s CRAY C90, every accomplish-ment brings the Department of Energy closer to its goal –more affordable, clean, and sustainable electricity.

— Holly A. Amundson

AcknowledgmentsFunding for the UAE wind turbine is provided by NREL’s

Chief of Applied Research, Mike Robinson, through theArmy/NASA CFD Code Project. C.P. Van Dam and students at the University of California at Davis providesupport for the computational component of NREL’sUAE project.

Editor’s Note: As Gridpoints went to press, Earl Duqueaccepted a new challenge at the College of Engineering andTechnology, Northern Arizona University in Flagstaff as anassociate professor of mechanical engineering. Duque is con-tinuing his research in wind turbine technology and appliedcomputational fluid dynamics.

A three-week test of the National RenewableEnergy Laboratory’s wind turbine was undertak-en in the world’s largest wind tunnel at NASAAmes Research Center. Researchers use aerody-namic and structural dynamic data collectedduring tests in the 80- by 120-foot tunnel, andnumerical simulations run on the CRAY C90computer at the NAS Facility to improve theunderstanding of wind turbines. This knowledgehelps engineers design and operate more cost-effective wind turbines. (NASA)

For the past six years,wind energy capacityhas increased at an

annual rate of 40 per-cent. The Wind Force 10report, prepared by theEuropean Wind Energy

Association (EWEA),predicts a 10 percent

increase in the world’swind energy productionby 2020. Ten percent, or1.2 million megawatts,

are equivalent toEurope’s electricity con-sumption for an entireyear. EWEA believes

wind energy expansionwill create 1.7 million

new jobs.

gp

18 Gridpoints

During the 1970s, researchers were constructing anetwork of computers to share information in aneffort to expand science. The resulting computer

network of the ’70s became the Internet of the 1980s and’90s. Today, researchers led by the NAS Systems Divisionare expanding the Internet model to build NASA’s Infor-mation Power Grid (IPG) – a distributed system linkingremotely located computer networks, storage devices, andinstruments into a single computational grid.

NASA has selected Ames as the Center of Excellence forInformation Technology and the NAS Systems Division asthe focal point for high performance computing as well asthe development of NASA’s prototype IPG. In September,NASA researchers will demonstrate the IPG’s capability andits ability to access geographically distributed systems.NASA’s IPG team will deliver a “persistent testbed” for sci-entists to test their applications, and to provide the frame-work for systems software development. To meet theSeptember 30 milestone, the team is developing more thana dozen tools to expand the grid’s infrastructure.

IPG’s Network ConnectivityThe first measurable step in the grid’s development wastaken last April when NAS researchers demonstrated theIPG network’s connectivity and its ability to access distrib-uted archival data. The April milestone sought a data trans-fer rate between 10 and 50 megabits per second whileretrieving files from two remote sites. Using NREN (theNASA Research and Education Network) for the connec-tion, the tests delivered a connectivity rate of 58 megabitsper second with data transfers from four separate locations.

“While the April demonstration is by no means high speed,network connectivity lays the foundation for the next step –a high-speed interconnect test,” says Leigh Ann Tanner,manager, advanced systems development.

‘Launch Pad’ is Gateway to the GridThe most visible piece of the grid’s infrastructure is theweb-based user interface, or portal, called the NASA IPG“Launch Pad.” This interface is being developed in collabo-ration with researchers at San Diego Supercomputer Center,the National Computational Science Alliance (NCSA),Argonne National Laboratory, and the University ofIndiana. The Launch Pad will allow researchers to createand submit a job, track its progress, and determine whichmachine is processing the work. This simple and easy-to-useweb interface will allow a user to specify the resourcesrequired to run an experiment, such as the amount ofmemory required and the number of computer processors,and then “launch” the job to the grid.

“Scientists often perform the same basic task repeatedly,”says George Myers, NAS scientific consultant and portaldevelopment group lead. “Being able to upload previouslycreated job templates will only require changes to a line ortwo of code, or the addition of new file specifications, with-out completely rebuilding job submission information eachtime. We are working to make the job template modular sothat researchers can customize their view of the portal.”

The portal group is finalizing the look and feel of the NASALaunch Pad, which will use a grid portal development kitadapted from NCSA’s Jason Novotny and Vaughn Welch’s

Infrastructure For NASA’sInformation Power Grid

Nears Completion

More than a dozen components of the IPG arebeing readied for roll-out by September 30.

Gridpoints 19www.nas.nasa.gov/gridpoints

interface to the grid. Dennis Gannon of the University ofIndiana has also been advising the portal development group.

User Support, Middleware Under DevelopmentTo answer grid users’ questions and troubleshoot problems,NASA is developing a new grid support model that willprovide 24 hour, seven day per week assistance to users,regardless of an end-user’s location. The support modelteam is tasked with developing procedures for problemreporting and tracking, routing questions to the propersupport group and how the situation will be remedied.

Varying levels of support exist at the three participatingNASA centers, ranging from phone or e-mail support onlyto the NAS Systems Division’s “24 x 7” assistance. For moreextensive support, the NAS scientific consultants are avail-able Monday through Friday from 7 a.m. to 5:30 p.m. (PST).Answers requiring other specific knowledge areas may residewith support staff located at other participating NASAresearch centers – currently Glenn and Langley. To addressthe intricacies of user assistance, the IPG team is develop-ing one mechanism to standardize all of the support func-tions that will be in place by September 30.

NASA’s prototype IPG will use the Globus middlewaretoolkit – developed by Ian Foster of Argonne NationalLaboratory and Carl Kesselman of the University of South-ern California’s Information Sciences Institute, to enabledifferent types of applications and interfaces to communi-cate with the grid. Judith Utley, a NASA IPG team mem-ber, coordinates upgrades and new releases of Globus toeach of the grid sites, and resolves software bugs. Utley also

manages the deployment of the GridInformation Service (GIS) thatallows researchers to poll the GIS todetermine what jobs are running, aswell as types of systems and proces-sors are available. “Globus deploy-ment and grid information servicesrequire close coordination betweenall NASA IPG centers,” says Utley.“Our weekly teleconferences providethe avenue to complete this phase ofthe project.” Utley collaborates onGIS deployment with GlennResearch Center’s Kim Johnson andAllen Holtz as well as LangleyResearch Center’s Mike Perry andGreg Cates.

Keeping the Grid Secure“Security for both the user and thegrid is of the utmost importance,”says NAS systems engineer YvonneMalloy, who is implementing theIPG’s certificate infrastructure usingNetscape CMS (Certificate

Management System) to address this issue. To obtainauthorization to run jobs on the IPG, a user will request acertificate through the online grid registration authority.Once a user has been authenticated and authorized, a cer-tificate is issued and users can download it onto their webbrowser. Each job is “signed” with the certificate. Malloy iscurrently incorporating the certificate system into theGlobus structure, and is planning to take user security tothe next level using the ‘MyProxy’ software system alsodeveloped by Welch and Novotny.

“MyProxy is a way for users to limit the exposure of theirprivate key password,” says Malloy. “The private key isdecrypted long enough to generate proxies and then it is re-encrypted. These proxies are valid for short amounts oftime, but are treated like the user’s actual certificate.”MyProxy is expected to be installed by the end of August.

Managing the ResultsAfter running a job on the IPG, results may be stored in alocation capable of handling tremendous volumes of data.For data management, NAS engineers are using a mass stor-age system for NASA’s IPG. Known as the Data MigrationFacility (DMF), it consists of two SGI Origin-based sys-tems, a large fiber channel RAID (Redundant Array ofInexpensive Disks) array capable of storing 1.3 terabytes ofdata, and four attached tape drives. The tape drives willmigrate stored files from the RAID array to removable tapecartridges. “Off-loading a file’s data blocks from the RAIDarray to tape will give us virtually unlimited storage capaci-ty,” says Bill Arasin, a systems analyst in the high speedprocessor group. “There will be a brief delay when users

NASA’s Information Power Grid links high performance computers at Ames, Glenn, andLangley Research Centers. Infrastructure enabling researchers to test applications anddevelop grid related systems software will be completed by September 30.

20 Gridpoints

retrieve their files, but this offers the abili-ty to store as much data as we have tape.”

The DMF is a redundant system thatmaintains a copy of each file on two sepa-rate disks in the event of a disk failure. Toadd further redundancy, both systems willbe configured to run Failsafe DMF whichenables either system to take over all theservices of the other if a system outageoccurs. “Failsafe DMF provides a seamlessshift between servers with no interruptionof service detectable by users,” says Arasin.

Proving the Grid’s ReliabilityTo ensure that IPG users will have a reli-able computing environment, NASAengineers have been testing the grid’s per-formance. Validation encompasses fourareas: Each piece of the grid’s infrastruc-ture is repeatedly tested to ensure that itworks and will perform adequately; that itwill interface with other systems and soft-ware; that it will perform uninterruptedwhile handling a large number of jobs; andthat it remains stable as more computa-tions are added into the system. “Our testsare designed to model the tasks that willmost frequently be done by users on a day-to-day basis inthe grid environment,” says Ray Turney, IPG test engineer.

“All building blocks of the grid’s infrastructure are interde-pendent upon the others,” Turney continues. “For example,when new software versions are introduced, the systemenvironment may change.” Nightly tests run simulated jobsthrough the systems at Glenn, Langley, and the localmachines at Ames. Turney is currently testing the Globusenvironment and the Metacomputing Directory Service.

Condor: Adding Flexibility to the IPG In addition to high performance computing, where jobspeed is measured in floating point operations per second,researchers can also use high throughput computing re-sources where large numbers of processors work for longperiods of time – measured in weeks or even months.“We’re not only linking supercomputers,” says Eric Lang-hirt, NAS Condor systems specialist, “but workstations thatare idle will now become a resource for the grid.” Condor,developed by Miron Livny and a team of researchers at theUniversity of Wisconsin, is a distributed batch processingsystem that scavenges compute cycles from idle workstationson a network and runs jobs while the local user is away.

The software system works by monitoring the CPU loadand keyboard activity of a pool of workstations to deter-mine whether a terminal is idle and available to run jobs.The system matches a job’s requirements for resources with

workstations that have available computing cycles. Once amatch is made, the job is sent to the workstation for pro-cessing. Condor continually monitors the workstation’sCPU load and its keyboard; movement of the mouse or akeystroke will restore control of the computer to the localuser. Condor then moves the job to the next idle worksta-tion in the pool and continues processing.

Using the Condor system, the NAS Facility’s more than400 workstations will provide nearly 2 million CPU hourseach year – equal in power to one of the NASA’s 256-processor high performance computers. When all threeNASA grid centers are linked, this number is expected tonearly double.

Looking AheadAdditional infrastructure components slated for completionthis fall include a distributed debugging and performanceanalysis system, a wide area network connection capable oftransmitting at least 100 megabits per second, a computerhours allocation and accounting system, and the integrationof the CORBA programming language to the IPG.

Each completed and integrated piece of the grid’s infra-structure brings the IPG team closer to meeting its Septem-ber goal of “demonstrating seamless operations within aheterogeneous distributed computing environment.”

— Nicholas A. Veronico

From left to right: Ames Center Director Henry McDonald, NAS Division Chief BillFeiereisen, NASA Administrator Daniel Goldin, and NAS senior scientist Jim Taftdiscuss high performance computing and system scalability in the NAS Facility’scomputer room. Lomax, the 512-processor SGI Origin 2800, shown in the back-ground, is a node on the IPG and was a key component in demonstrating thegrid’s connectivity and ability to access distributed archival data.

(Nicholas A. Veronico)

gp

September 11-15 – Albuquerque, New MexicoThe Linux Supercluster Users Conference is aimed at shar-ing information on scientific computing techniques. Theobjective of this meeting is to help computational scientistsand engineers develop applications to achieve maximumperformance and scalability on Linux Supercluster systems.Hosted by the National Computational Science Alliance(NCSA), the Advanced Computing Technology Center atIBM Research, and the High Performance ComputingEducation and Research Center. For additional informa-tion, visit:http://clumber.tc.cornell.edu/actc/Supercluster/index.html

September 19-20 – Mountain View, CaliforniaNASA’s Information Power Grid project will host a work-shop to demonstrate and review the IPG’s growth, examinecurrent issues, and plan future directions. Workshop pro-ceeding and selected papers will be published in a specialissue of “Scientific Programming.” Contact MarciaRedmond at (650) 604-4373 or visit the workshop’swebsite at: http://www.ipg.nasa.gov/ workshop.html

September 18-20 – Davis, CaliforniaThe Fifth Symposium on Overset Grids and SolutionTechnology will be held at UC Davis. This internationalsymposium provides a joint forum where mathematicians,scientists, and engineers from academia, industry, and fed-eral laboratories can exchange research ideas, new algorith-mic and software developments, and applications pertinentto the state-of-the-art and state-of-the-practice of Overset/Chimera Grid and Solution Technology. More informationabout the symposium can be found at:http://ntserver.itd.ucdavis.edu/Chimera2000/

September 28 – Mountain View, CaliforniaSGI will roll out its new Origin and Onyx 3000 modularcomputing systems at its Mountain View headquartersfrom 2 p.m. to 7 p.m. Bill Feiereisen, NAS SystemsDivision chief, is a featured guest speaker. For directions,registration, and additional information, call (800) 318-6432, or visit: http://www.sgi.com/events/thinkahead

October 9-13 – San Jose, CaliforniaThe Microprocessor Forum 2000 provides designers andusers of high-performance microprocessors with insightsinto new processor designs – for applications ranging fromPCs and workstations to information appliances and theInternet. For additional information, visit the conference’swebsite at: http://www.MDRonline.com/MPF/conf.html

October 30-November 1 – Orlando, FloridaStorage Networking World Conference and Expo featurespanel discussions and case studies on the future of StorageNetworking and topics such as storage outsourcing, man-aging storage and server assets, Storage Area Networkinteroperability, the role of storage systems within theInternet infrastructure and integrating diverse environ-ments, along with a separate track covering in-depth tech-nical issues and deployment examples associated with stor-age networks. More information is available at:http://www.storagenetworkingworld.com

November 3-5 – Bethesda, MarylandEighth Foresight Conference on Molecular Nanotech-nology. This conference is a meeting of scientists and tech-nologists working in fields leading toward molecular nan-otechnology: thorough three-dimensional structuralcontrol of materials and devices at the molecular level. Anintroductory tutorial on foundations of nanotechnologywill be held November 2. For more information, phone(650) 917-1122, or fax (650) 917-1123. The conference’sweb address is:http://www.foresight.org/Conferences/MNT8/index.html

November 4-10 – Dallas, TexasSC2000, the conference of high performance networkingand computing, will meet at the Dallas Convention Centerfor a series of technical programs, demonstrations, educa-tional outreach, and visualizations of computational data.SC2000 is sponsored by the Institute of Electrical andElectronics Engineers Computer Society and the Associa-tion for Computing Machinery’s Special Interest Group onComputer Architecture. Be sure to visit the NASA booth.Further details can be obtained at: http://www.sc2000.org

Gridpoints 21www.nas.nasa.gov/gridpoints

Calendar ofEvents

Mul

ti-w

all

nano

tube

com

pres

sion

(bu

cklin

g),

sim

ulat

ion

and

visu

aliz

atio

n by

Dee

pak

Sriv

asta

va

gp