1. Swift Fast, Reliable, Loosely Coupled Parallel Computation Ian Foster Computation Institute Argonne National Laboratory University of Chicago Joint work with Yong Zhao, Ioan Raicu, Mike Wilde, Ben Clifford, Mihael Hatigan, Tibi Stef-Praun, Veronika Nefedova

2. 2 Case Study: Functional MRI (fMRI) Data Center l Online repository of neuroimaging data l A typical study comprises 3 groups, 20 subjects/group, 5 runs/subject, 300 volumes/run  90,000 volumes, 60 GB raw  1.2 million files processed l 100s of such studies in total www.fmridc.org

3. 3 Many Users Analyze fMRI Data l Wide range of analyses u Testing, interactive analysis, production runs u Data mining u Parameter studies

4. 4 Three Obstacles to Creating a Community Resource l Accessing messy data u Idiosyncratic layouts & formats u Data integration a prerequisite to analysis l Implementing complex computations u Expression, discovery, reuse of analyses u Scaling to large data, complex analyses l Making analysis a community process u Collaboration on both data & programs u Provenance: tracking, query, application

5. 5 The Swift Solution (Or: Outline of this Talk) l Accessing messy data u Idiosyncratic layouts & formats u Data integration a prerequisite to analysis l Implementing complex computations u Expression, discovery, reuse of analyses u Scaling to large data, complex analyses l Making analysis a community process u Collaboration on both data & programs u Provenance: tracking, query, application XDTM SwiftScript Karajan +Falkon VDC

6. 6 The Swift Solution (Or: Outline of this Talk) l Accessing messy data u Idiosyncratic layouts & formats u Data integration a prerequisite to analysis l Implementing complex computations u Expression, discovery, reuse of analyses u Scaling to large data, complex analyses l Making analysis a community process u Collaboration on both data & programs u Provenance: tracking, query, application XDTM SwiftScript Karajan +Falkon VDC

7. 7 l Scientific data is often logically structured u E.g., hierarchical structure u Common to map functions over dataset members u Nested map operations can scale to millions of objects The Messy Data Problem (1)

8. 8 The Messy Data Problem (2) l Heterogeneous storage format & access protocols u Same dataset can be stored in text file, spreadsheet, database, … u Access via filesystem, DBMS, HTTP, WebDAV, … l Metadata encoded in directory and file names l Hinders program development, composition, execution./knottastic drwxr-xr-x 4 yongzh users 2048 Nov 12 14:15 AA drwxr-xr-x 4 yongzh users 2048 Nov 11 21:13 CH drwxr-xr-x 4 yongzh users 2048 Nov 11 16:32 EC./knottastic/AA: drwxr-xr-x 5 yongzh users 2048 Nov 5 12:41 04nov06aa drwxr-xr-x 4 yongzh users 2048 Dec 6 12:24 11nov06aa. /knottastic//AA/04nov06aa: drwxr-xr-x 2 yongzh users 2048 Nov 5 12:52 ANATOMY drwxr-xr-x 2 yongzh users 49152 Dec 5 11:40 FUNCTIONAL. /knottastic/AA/04nov06aa/ANATOMY: -rw-r--r-- 1 yongzh users 348 Nov 5 12:29 coplanar.hdr -rw-r--r-- 1 yongzh users 16777216 Nov 5 12:29 coplanar.img. /knottastic/AA/04nov06aa/FUNCTIONAL: -rw-r--r-- 1 yongzh users 348 Nov 5 12:32 bold1_0001.hdr -rw-r--r-- 1 yongzh users 409600 Nov 5 12:32 bold1_0001.img -rw-r--r-- 1 yongzh users 348 Nov 5 12:32 bold1_0002.hdr -rw-r--r-- 1 yongzh users 409600 Nov 5 12:32 bold1_0002.img -rw-r--r-- 1 yongzh users 496 Nov 15 20:44 bold1_0002.mat -rw-r--r-- 1 yongzh users 348 Nov 5 12:32 bold1_0003.hdr -rw-r--r-- 1 yongzh users 409600 Nov 5 12:32 bold1_0003.img

9. 9  XML Dataset Typing & Mapping (XDTM) l Describe logical structure by XML Schema u Primitive scalar types: int, float, string, date, … u Complex types (structs and arrays) l Use mapping descriptors for mappings u How dataset elements are mapped to physical representations u External parameters (e. g. location) l Use XPath for dataset selection XDTM: XML Dataset Typing and Mapping for Specifying Datasets [EGC05]

10. 10 XDTM: Related Work l Data format standardization u FITS, CDF, HDF-5, DICOM l Data format description u DFDL [Beckerle,Westhead04] embeds annotations with XML Schema u PADS [Fisher,Gruber05], PADX [Fernandez,Fisher06], declarative specs of physical layout & semantics l Logical object u ADO [Microsoft01], in-memory relational model u SDO [Beatty,Brodsky03], logical data model for J2EE

11. 11 XDTM: Implementation l Virtual integration u Each data source treated as virtual XML source u Data structure defined as XML schema u Mapper responsible for accessing source and translating to/from XML representation u Bi-directional l Common mapping interface u Data providers implement the interface l Responsible for data access details u Standard mapper implementations provided l String, file system, CSV, …

12. 12 The Swift Solution (Or: Outline of this Talk) l Accessing messy data u Idiosyncratic layouts & formats u Data integration a prerequisite to analysis l Implementing complex computations u Expression, discovery, reuse of analyses u Scaling to large data, complex analyses l Making analysis a community process u Collaboration on both data & programs u Provenance: tracking, query, application XDTM SwiftScript Karajan +Falkon VDC

13. 13 SwiftScript l Typed parallel programming notation u XDTM as data model and type system u Typed dataset and procedure definitions l Scripting language u Implicit data parallelism u Program composition from procedures u Control constructs (foreach, if, while, …) Clean application logic Type checking Dataset selection, iteration Discovery by types Type conversion A Notation & System for Expressing and Executing Cleanly Typed Workflows on Messy Scientific Data [SIGMOD05]

14. 14 SwiftScript: Related Work l Coordination language u Linda [Ahuja,Carriero86], Strand [Foster,Taylor90], PCN [Foster92] u Durra [ Barbacci,Wing86 ], MANIFOLD [Papadopoulos98] u Components programmed in specific language (C, FORTRAN) and linked with system l “Workflow” languages and systems  Taverna [Oinn,Addis04], Kepler [Lud ä scher,Altintas05], Triana [Churches,Gombas05], Vistrail [Callahan,Freire06], DAGMan, Star-P u XPDL [WfMC02], BPEL [Andrews,Curbera03], and BPML [BPML02], YAWL [van de Aalst,Hofstede05], Windows Workflow Foundation [Microsoft05]

15. 15 Related Work SwiftScriptBPELXPDLMW WflowDAGManTavenaTrianaKeplerVistrailStar-P Scales to Grids ++--- ----+ Typing ++ ---+-+ Iteration ++-/+-+---+-+ Scripting ++--+++--+ Dataset Mapping +--------- Service Interop +-+----+-- Subflow/comp. +-++--++-+ Provenance +--+-+-++- Open source +++-+++++- “A 4x200 flow leads to a 5 MB BPEL file … chemists were not able to write in BPEL” [Emmerich,Buchart06]

16. 16 fMRI Type Definitions in SwiftScript type Study { Group g[ ]; } type Group { Subject s[ ]; } type Subject { Volume anat; Run run[ ]; } type Run { Volume v[ ]; } type Volume { Image img; Header hdr; } type Image {}; type Header {}; type Warp {}; type Air {}; type AirVec { Air a[ ]; } type NormAnat { Volume anat; Warp aWarp; Volume nHires; } Simplified version of fMRI AIRSN Program (Spatial Normalization)

17. 17 Type Definitions in XML Schema <xs:schema targetNamespace="http://www.fmri.org/schema/airsn.xsd" xmlns="http://www.fmri.org/schema/airsn.xsd" xmlns:xs="http://www.w3.org/2001/XMLSchema">

18. 18 fMRI Example: AIRSN Program Definition (Run snr) functional ( Run r, NormAnat a, Air shrink ) { Run yroRun = reorientRun( r, "y" ); Run roRun = reorientRun( yroRun, "x" ); Volume std = roRun[0]; Run rndr = random_select( roRun, 0.1 ); AirVector rndAirVec = align_linearRun( rndr, std, 12, 1000, 1000, "81 3 3" ); Run reslicedRndr = resliceRun( rndr, rndAirVec, "o", "k" ); Volume meanRand = softmean( reslicedRndr, "y", "null" ); Air mnQAAir = alignlinear( a.nHires, meanRand, 6, 1000, 4, "81 3 3" ); Warp boldNormWarp = combinewarp( shrink, a.aWarp, mnQAAir ); Run nr = reslice_warp_run( boldNormWarp, roRun ); Volume meanAll = strictmean( nr, "y", "null" ) Volume boldMask = binarize( meanAll, "y" ); snr = gsmoothRun( nr, boldMask, "6 6 6" ); } (Run or) reorientRun (Run ir, string direction) { foreach Volume iv, i in ir.v { or.v[i] = reorient(iv, direction); } Collaboration with James Dobson, Dartmouth [SIGMOD05]

19. 19 Expressiveness Lines of code with different encodings Collaboration with James Dobson, Dartmouth [SIGMOD05] ApplnScriptGenerator Swift Script ATLAS149726 ATLAS29713510 FILM16313417 FEAT8419113 AIRSN215~40034

20. 20 Expressiveness Lines of code with different encodings ApplnScriptGenerator Swift Script ATLAS149726 ATLAS29713510 FILM16313417 FEAT8419113 AIRSN215~40034 Collaboration with James Dobson, Dartmouth [SIGMOD05]

21. 21 The Swift Solution (Or: Outline of this Talk) l Accessing messy data u Idiosyncratic layouts & formats u Data integration a prerequisite to analysis l Implementing complex computations u Expression, discovery, reuse of analyses u Scaling to large data, complex analyses l Making analysis a community process u Collaboration on both data & programs u Provenance: tracking, query, application XDTM SwiftScript Karajan +Falkon VDC

22. 22 Swift Runtime System l Runtime system for SwiftScript u Translate programs into task graphs u Schedule, monitor, execute task graphs on local clusters and/or distributed Grid resources u Annotate data products with provenance metadata l Grid scheduling and optimization u Lightweight execution engine: Karajan u Falkon: lightweight dispatch, dynamic provisioning u Grid execution: site selection, data movement u Caching, pipelining, clustering, load balancing u Fault tolerance, exception handling A Virtual Data System for Representing, Querying & Automating Data Derivation [SSDBM02]

23. 23 Swift Runtime: Related Work l Multi-level scheduling [Banga99,Stankovic99] l Condor glidein [Frey02], Condor Brick [Singh05,Mehta06], MyCluster [Walker06] l Adaptive resource control [Appleby01], [Ramakrishnan06] l Lightweight dispatch [Anderson04]

24. 24 Virtual Node(s) SwiftScript Abstract computation Virtual Data Catalog SwiftScript Compiler SpecificationExecution Virtual Node(s) Provenance data Provenance data Provenance collector launcher file1 file2 file3 App F1 App F2 Scheduling Execution Engine (Karajan w/ Swift Runtime) Swift runtime callouts C CCC Status reporting Swift Architecture Provisioning Falkon Resource Provisioner Amazon EC2

25. 25 Swift uses Karajan Workflow Engine l Fast, scalable threading model l Suitable constructs for control flow l Flexible task dependency model u “Futures” enable pipelining l Flexible provider model allows for use of different run time environments u Job execution and data transfer u Flow controlled to avoid resource overload l Workflow client runs from a Java container Java CoG Workflow, Gregor von Laszewski, Mihael Hatigan, 2007

26. 26 Karajan Futures Enable Pipelining (Dispatch is performed here via GRAM+PBS)

27. 27 Karajan Futures Enable Pipelining (Dispatch is performed here via GRAM+PBS)

28. 28 Execution Virtual Node(s) Swift Can Use Falkon Dispatcher & Provisioner Provisioning Falkon Resource Provisioner Amazon EC2 Reqs l Falkon provisioner: u Monitors demand (incoming user requests) u Manages supply: selects resources; creates executors (via Globus GRAM) u Various decision strategies for acquisition and release l Falkon executor u Streamlined task dispatch u Driven by Karajan l Dispatch to other executors also supported—e.g., GRAM Ioan Raicu, U.Chicago Falkon Executor.......

29. Falkon Dispatch: Throughput, Scalability Ioan Raicu, U.Chicago

30. 30 Falkon Provisioning: Synthetic Benchmark 18 Stages 1000 tasks 17820 CPU seconds 1260 total time on 32 machines Ioan Raicu & Yong Zhao, U.Chicago

31. Release after 15 Seconds Idle Ioan Raicu & Yong Zhao, U.Chicago

32. Release after 180 Seconds Idle Ioan Raicu & Yong Zhao, U.Chicago

33. 33 Swift Application Performance: fMRI Task Graph Yong Zhao and Ioan Raicu, U.Chicago

34. 34 Swift Application Performance: fMRI Task Graph Yong Zhao and Ioan Raicu, U.Chicago

35. Swift Application B. Berriman, J. Good (Caltech) J. Jacob, D. Katz (JPL)

36. 36 Montage Yong Zhao and Ioan Raicu, U.Chicago

37. 37 Other Swift Applications Include … Application#Jobs/computationLevels ATLAS* HEP Event Simulation500K1 fMRI AIRSN Image Processing100s12 FOAM* Ocean/Atmosphere Model2000 (250 8-CPU jobs)3 GADU* Genomics: (14M seq. analyzed)40K4 fMRI Aphasia Study5004 NVO/NASA Montage1000s16 QuarkNet/I2U2* + Physics Science Education10s3-6 RadCAD: Radiology Classifier Training1000s5 SIDGrid: EEG Wavelet Proc, Gaze Analysis, …100s20 SDSS* Coadd, Cluster Search40K, 500K2, 8 * Using predecessor Virtual Data System (VDS) + Collaborative science learning & education: 18 experiments, 51 universities/labs, 500+ schools, 100,000 students

38. 38 The Swift Solution (Or: Outline of this Talk) l Accessing messy data u Idiosyncratic layouts & formats u Data integration a prerequisite to analysis l Implementing complex computations u Expression, discovery, reuse of analyses u Scaling to large data, complex analyses l Making analysis a community process u Collaboration on both data & programs u Provenance: tracking, query, application XDTM SwiftScript Karajan +Falkon VDC

39. 39 Virtual Data Concept l Capture information about relationships among u Data (varying locations and representations) u Programs (& inputs, outputs, constraints) u Computations (& execution environments) l Apply this information to: u Discovery of data and programs u Computation management u Provenance u Planning and scheduling u Performance optimization Data Program Computation operates-on execution-of created-by consumed-by A Virtual Data System for Representing, Querying & Automating Data Derivation [SSDBM02]

40. 40 Provenance – Related Work l Database u Determine the source of tuples [Cui,Widom00] u Why and where [Buneman,Khanna01] l Scientific u Logbook [Myers,Chappell03] [Bourilkov,Khandelwal06] l Service u P-assertions [Szomszor,Moreau03] l Other u PASS [Muniswamy-Reddy,Holland06]

41. 41 Provenance Model l Temporal aspect u Prospective provenance l Recipes for how to produce data l Metadata annotations about procedures and data u Retrospective provenance [GCE06] l Invocation records of run time environments and resources used: site, host, executable, execution time, file stats... l Dimensional aspect u Virtual data relationships u Derivation lineage u Metadata annotations Relationships Derivation Lineage Metadata Annotations Applying the Virtual Data Provenance Model [IPAW06]

42. 42 Virtual Data Schema

43. 43 Query Context: fMRI Analysis First Provenance Challenge, http://twiki.ipaw.info/ [CCPE06]

44. 44 Query Examples l Query by procedure signature u Show procedures that have inputs of type subjectImage and output types of warp l Query by actual arguments u Show align_warp calls (including all arguments), with argument model=rigid l Query by annotation u List anonymized subject images for young subjects: l Find datasets of type subjectImage, annotated with privacy=anonymized and subjectType=young l Basic lineage graph queries u Find all datasets derived from dataset ‘5a’ l Graph pattern matching u Show me all output datasets of softmean calls that were aligned with model=affine l Multi-dimensional query

45. 45 Acknowledgements l Swift effort is supported by NSF (I2U2, iVDGL), NIH, UChicago/Argonne Computation Institute l Swift team u Ben Clifford, Ian Foster, Mihael Hategan, Veronika Nefedova, Ioan Raicu, Mike Wilde, Yong Zhao l Java CoG Kit u Mihael Hategan, Gregor Von Laszewski, and many collaborators l User contributed workflows and application use u I2U2, ASCI Flash, U.Chicago Molecular Dynamics, U.Chicago Radiology, Human Neuroscience Lab

46. 46 Future Work l XDTM u Support for services as well as applications u Greater abstraction in mappers; databases l SwiftScript u Exceptions u Event-driven dispatch & execution l Falkon u Scale to more resources; data caching u Support for service workloads l VDC u Integration into Swift; collaboration support u Experiments at scale

47. 47 Swift: Summary l Clean separation of logical/physical concerns u XDTM specification of logical data structures + Concise specification of parallel programs u SwiftScript, with iteration, etc. + Efficient execution (on distributed resources) u Karajan+Falkon: Grid interface, lightweight dispatch, pipelining, clustering, provisioning + Rigorous provenance tracking and query u Virtual data schema & automated recording  Improved usability and productivity u Demonstrated in numerous applications http://www.ci.uchicago.edu/swift

48. 48 Thank You!

49. Extra Slides

50. 50 XDTM – Related Work Features\ModelXDTMDFDLPADS/PADXSDO Data FormatAny Binary file format Ad hoc data source Any Abstract Data Model Declarative description XML Schema Declarative description XML Schema Disconnected data graph Physical Information Mapping descriptor Annotation embedded in XML schema Mixed with declaration N/A Logical/Physical Separation YesNo Yes Mapping Specification Mapping interface No Data Access Service

51. 51 VDS – System Diagram VDL Program Virtual Data Catalog Workflow Generator Abstract Workflow Planner Execution Plan Workflow Enactor Provenance Collector Launcher Grid Provider

52. 52 Application – CMS Event Analysis mass = 200 decay = WW stability = 1 event = 8 mass = 200 decay = WW stability = 1 plot = 1 mass = 200 decay = WW plot = 1 mass = 200 decay = WW event = 8 mass = 200 decay = WW stability = 1 mass = 200 decay = WW stability = 3 mass = 200 decay = WW mass = 200 decay = ZZ mass = 200 decay = bb mass = 200 plot = 1 mass = 200 event = 8 A e-workspace of simulated data is created for future use by scientists... Collaboration with Rick Cavanaugh, Dimitri Bourilkov et al. University of Florida [CHEP03]

53. 53 ATLAS Large Scale Simulation “How much compute time was delivered?” (BNL, 1475+ CPUs) CPU yearsJobsMonth 0.4514022004-06 19.88132672004-07 33.88206782004-08 40.06202292004-09 15.21308332004-10 14.95345912004-11

54. 54 Fast Ocean Atmosphere Model NCAR Manual config, execution, bookkeeping VDS on Teragrid Automated Visualization courtesy Pat Behling and Yun Liu, UW Madison

55. 55 Radiology Example type Image {} type Center {} type ROI { Image image; Center center; } type ROIVec { ROI roi[]; } (AzInfo az) LDAClassify (ROIVec malROIs, ROIVec benROIs, Parameter param, FeatureNames fn) { …. } ROIVec malROIs ; ROIVec benROIs ; Parameter param ; FeatureNames featureList ; AzInfo az ; az = LDAClassify(malROIs, benROIs, param, featureList);

56. 56 An International Community Data source: since Oct. 2003, http://www.fmridc.org

57. 57 Workflow Status and logs swift command launcher f1 f2 f3 Worker Nodes App a1 App a2 Using Swift SwiftScript App a1 App a2 Data f1 f2 f3 site list app list Provenance data

58. 58 Virtual Data Concept (1) l Capture information about relationships among u Data (varying locations and representations) u Programs (& inputs, outputs, constraints) u Computations (& execution environments) l Apply this information to: u Discovery of data and programs u Computation management u Provenance u Planning and scheduling u Performance optimization Data Program Computation operates-on execution-of created-by consumed-by A Virtual Data System for Representing, Querying & Automating Data Derivation [SSDBM02]

59. 59 Virtual Data Concept (2) l Location transparency u Data processing independent of location u Replica location service, selection service l Materialization transparency u Recipes for data derivation l Physical representation transparency u Logical descriptions and relations A Virtual Data System for Representing, Querying & Automating Data Derivation [SSDBM02]

60. 60 Virtual Data System (VDS) l Introduced Virtual Data Language (VDL) u A location-independent parallel language l Several planners, e.g.: u Pegasus: main production planner u Euryale: experimental “just in time” planner u GADU/GNARE: user application planner (D. Sulahke, Argonne) l Provenance u Kickstart: app launcher and tracker u VDC: virtual data catalog A Virtual Data System for Representing, Querying & Automating Data Derivation [SSDBM02]

61. 61 VDL/VDS Limitations l Missing language features u Data typing & data mapping u Iterators & control-flow constructs l Run time complexity in VDS u State explosion for data-parallel applications u Computation status hard to provide u Debugging information complex & distributed l Performance u Still many runtime bottlenecks A Virtual Data System for Representing, Querying & Automating Data Derivation [SSDBM02]

62. 62 Swift Application Example: ACTIVAL: Neural Activation Validation Identifies clusters of neural activity not likely to be active by random chance: switch labels of the conditions for one or more participants; calculate the delta values in each voxel, re-calculate the reliability of delta in each voxel, and evaluate clusters found. If the clusters in data are greater than the majority of the clusters found in the permutations, then the null hypothesis is refuted indicating that clusters of activity found in our experiment are not likely to be found by chance. Work by S. Small, U. Hasson, UChicago.

63. 63 SwiftScript Program ACTIVAL – Datatypes & Utilities type script {} type fullBrainData {} type brainMeasurements{} type fullBrainSpecs {} type precomputedPermutations{} type brainDataset {} type brainClusterTable {} type brainDatasets{ brainDataset b[]; } type brainClusters{ brainClusterTable c[]; } // Procedure to run "R" statistical package (brainDataset t) bricRInvoke (script permutationScript, int iterationNo, brainMeasurements dataAll, precomputedPermutations dataPerm) { app { bricRInvoke @filename(permutationScript) iterationNo @filename(dataAll) @filename(dataPerm); } } // Procedure to run AFNI Clustering tool (brainClusterTable v, brainDataset t) bricCluster (script clusterScript, int iterationNo, brainDataset randBrain, fullBrainData brainFile, fullBrainSpecs specFile) { app { bricPerlCluster @filename(clusterScript) iterationNo @filename(randBrain) @filename(brainFile) @filename(specFile); } } // Procedure to merge results based on statistical likelhoods (brainClusterTable t) bricCentralize ( brainClusterTable bc[]) { app { bricCentralize @filenames(bc); } }

64. 64 ACTIVAL: Dataset Iteration Procedures // Procedure to iterate over the data collection (brainClusters randCluster, brainDatasets dsetReturn) brain_cluster (fullBrainData brainFile, fullBrainSpecs specFile) { int sequence[]=[1:2000]; brainMeasurements dataAll ; precomputedPermutations dataPerm ; script randScript ; script clusterScript ; brainDatasets randBrains ; foreach int i in sequence { randBrains.b[i] = bricRInvoke(randScript,i,dataAll,dataPerm); brainDataset rBrain=randBrains.b[i]; (randCluster.c[i],dsetReturn.b[i]) = bricCluster(clusterScript,i,rBrain, brainFile,specFile); }

65. 65 ACTIVAL: Main Program // Declare datasets fullBrainData brainFile ; fullBrainSpecs specFile ; brainDatasets randBrain ; brainClusters randCluster<simple_mapper; prefix="Tmean.4mm.perm", suffix="_ClstTable_r4.1_a2.0.1D">; brainDatasets dsetReturn<simple_mapper; prefix="Tmean.4mm.perm", suffix="_Clustered_r4.1_a2.0.niml.dset">; brainClusterTable clusterThresholdsTable ; brainDataset brainResult ; brainDataset origBrain ; // Main program – executes the entire application (randCluster, dsetReturn) = brain_cluster(brainFile, specFile); clusterThresholdsTable = bricCentralize (randCluster.c); brainResult = makebrain(origBrain,clusterThresholdsTable,brainFile,specFile);

66. 66 ATLAS Event Simulation Sep 10 Mid July CPU-day

67. 67 Sloan Digital Sky Survey Sky region processed450 sq deg Num of sky fields14,000 Num of procedure calls5,000 Num of files120,000 Num of Grid sites4 Num of Processors used500 total Galaxy cluster identified60,000 5days vs. 10 months for the whole dataset (7000 sq deg)

68. 68 Web Interface Collaboration with Marge Bardeen, Tom Jordan, Liz Quigg, Eric Gilbert, Paul Nepywoda, Fermilab [CCGRID05] [FGCS05]

69. 69 Sample Program

70. 70 Web Services l WS described/imported as procedures l Dynamic invocation l XSLT as glue