1. Agenda 1. Background: INRIA, ActiveEon 2. ProActive Open Source: Programming, Scheduling, Resourcing 3. Cloud Seeding with GPU 4. UC: Genomics, Finance, IT 4. ProActive PACA GRID: Cloud Portal with GPUs D. Caromel, et al. Scientific Grid and Cloud Portal with ProActive Parallel Suite Portals are Complex: Hiding Infra. Details + Seamless Integration with Cloud, Multi-Cores, GPU

2. 2 2 1. Background

3. 3 3 OASIS Team Composition (35)  Researchers (5):  D. Caromel (UNSA, Det. INRIA)  E. Madelaine (INRIA)  F. Baude (UNSA)  F. Huet (UNSA)  L. Henrio (CNRS)  PhDs (11):  Antonio Cansado (INRIA, Conicyt)  Brian Amedro (SCS-Agos)  Cristian Ruz (INRIA, Conicyt)  Elton Mathias (INRIA-Cordi)  Imen Filali (SCS-Agos / FP7 SOA4All)  Marcela Rivera (INRIA, Conicyt)  Muhammad Khan (STIC-Asia)  Paul Naoumenko (INRIA/Région PACA)  Viet Dung Doan (FP6 Bionets)  Virginie Contes (SOA4ALL)  Guilherme Pezzi (AGOS, CIFRE SCP)  + Visitors + Interns  PostDoc (1):  Regis Gascon (INRIA)  Engineers (10):  Elaine Isnard (AGOS)  Fabien Viale (ANR OMD2, Renault )  Franca Perrina (AGOS)  Germain Sigety (INRIA)  Yu Feng (ETSI, FP6 EchoGrid)  Bastien Sauvan (ADT Galaxy)  Florin-Alexandru.Bratu (INRIA CPER)  Igor Smirnov (Microsoft)  Fabrice Fontenoy (AGOS)  Open position (Thales)  Trainee (2):  Etienne Vallette d’Osia (Master 2 ISI)  Laurent Vanni (Master 2 ISI)  Assistants (2):  Patricia Maleyran (INRIA)  Sandra Devauchelle (I3S) Located in Sophia Antipolis, between Nice and Cannes, Visitors Welcome!

4. 4 4 OASIS Team & INRIA: Parallelism, Cloud GPU nodes

5. 5 5  Co-developing, Support for ProActive Parallel SuiteProActive Parallel Suite  Worldwide Customers: Fr, UK, Boston USA Startup Company Born of INRIA Some Partners: Some Customers:

6. 6 2. ProActive Parallel Suite 6

7. 7 Product: ProActive Parallel Suite Java Parallel Toolkit Multi-Platform Job Scheduler Resource Manager Strong Differentiation:  Java Parallel Programming + Integration +  Portability: Linux, Windows, Mac +  Versatility: Desktops, Cluster, Grid, Clouds = Perfect Flexibility Used in Production Today: 50 Cores  300 Cores early 2010

8. 8 8 ProActive Programming View 8 GPU nodes

9. 9 9 ProActive Programming: Active Objects

10. 10 Broadcast and Scatter JVM ag cg ag.bar(cg); // broadcast cg ProActive.setScatterGroup(cg) ; ag.bar(cg); // scatter cg c1 c2 c3 c1 c2 c3 c1 c2 c3 c1 c2 c3 c1 c2 c3 c1 c2 c3 s c1 c2 c3 s Broadcast is the default behavior Use a group as parameter, Scattered depends on rankings

11. 11 Dynamic Dispatch Group JVM ag cg c1 c2 c3 c4 c5 c6 c7 c8c0 c9c1 c2 c3 c4 c5 c6 c7 c8c0 c9 c1 c2 c3 c4 c5 c6 c7 c8c0 c9 Slowest Fastest ag.bar(cg);

12. Abstractions for Parallelism The right Tool to do the Task right

13. 13 ProActive Parallel Suite  Workflows in Java  Master/Workers  SPMD  Components  …

14. 14 IC2D: Optimizing

15. 15 IC2D

16. 16 IC2D

17. 17 ChartIt

18. 18 Pies for Analysis and Optimization

19. 19 Video 1: IC2D Optimizing Monitoring, Debugging, Optimizing

20. 20 Scheduling & Resourcing

21. 21 ProActive Scheduling 21

22. 22 ProActive Scheduling Big Picture RESOURCES  Multi-platform Graphical Client (RCP)  File-based or LDAP authentication  Static Workflow Job Scheduling, Native and Java tasks, Retry on Error, Priority Policy, Configuration Scripts,…  Dynamic and Static node sources, Resource Selection by script, Monitoring and Control GUI,…  ProActive Deployment capabilities: Desktops, Clusters, Clouds,… ProActive Scheduler ProActive Scheduler ProActive Resource Manager ProActive Resource Manager

23. 23 Job Workflow Example : Picture Denoising Split Denoise Merge with selection on native executable availability (ImageMagik, GREYstoration) Multi-platform selection and command generation with file transfer in pre/post scripts

24. 24 ProActive Resourcing 24

25. 25 RESOURCING User Interface 25

26. 26 Clusters to Grids to Clouds: e.g. on Amazon EC2

27. 27 Node source Usecase : Configuration for external cloud with EC2 ProActive Scheduler ProActive Scheduler ProActive Resource Manager ProActive Resource Manager Dedicated resources LSF Static Policy Amazon EC2 EC2 Dynamic Workload Policy Desktops Timing Policy 12/24

28. 28 Video 2: Scheduler, Resource Manager

29. 3. Cloud Seeding with GPU 29

30. 30 Cloud Seeding with ProActive  Amazon EC2 Execution  Cloud Seeding strategy to mix heterogeneous computing resources :  External GPU resources

31. 31 Amazon EC2 GPU nodes CPU nodes ProActive Scheduler + Resource Manager Web Interface User Noised video file Cloud Seeding with ProActive

32. 32 Amazon EC2 GPU nodes CPU nodes ProActive Scheduler + Resource Manager Web Interface User User submit its noised video to the web interface Cloud Seeding with ProActive

33. 33 Amazon EC2 GPU nodes CPU nodes ProActive Scheduler + Resource Manager Web Interface User Web Server submit a denoising job the ProActive Scheduler Cloud Seeding with ProActive

34. 34 Amazon EC2 GPU nodes CPU nodes ProActive Scheduler + Resource Manager Web Interface User CPU nodes are used to split the video into smaller ones Cloud Seeding with ProActive

35. 35 Amazon EC2 GPU nodes CPU nodes ProActive Scheduler + Resource Manager Web Interface User CPU nodes are used to split the video into smaller ones Cloud Seeding with ProActive

36. 36 Amazon EC2 GPU nodes CPU nodes ProActive Scheduler + Resource Manager Web Interface User GPU nodes are responsible to denoise these small videos Cloud Seeding with ProActive

37. 37 Amazon EC2 GPU nodes CPU nodes ProActive Scheduler + Resource Manager Web Interface User GPU nodes are responsible to denoise these small videos Cloud Seeding with ProActive

38. 38 Amazon EC2 GPU nodes CPU nodes ProActive Scheduler + Resource Manager Web Interface User CPU nodes merge the denoised video parts Cloud Seeding with ProActive

39. 39 Amazon EC2 GPU nodes CPU nodes ProActive Scheduler + Resource Manager Web Interface User CPU nodes merge the denoised video parts Cloud Seeding with ProActive

40. 40 Amazon EC2 GPU nodes CPU nodes ProActive Scheduler + Resource Manager Web Interface User The final denoised video is sent back to the user Cloud Seeding with ProActive

41. 41 4.: Use Case 1: Genomics

42. 42 Resources set up Cluster Desktops Clouds EC2 SOLID machine from Nodes can be dynamically added! 16 nodes

43. 43 First Benchmarks  The distributed version with ProActive of Mapreads has been tested on the INRIA cluster with two settings: the Reads file is split in either 30 or 10 slices  Use Case: Matching 31 millions Sequences with the Human Genome (M=2, L=25) 4 Time FASTER from 20 to 100 Speed Up of 80 / Th. Sequential : 50 h  35 mn On going Benchmarks on Windows Desktops and HPCS 2008 … EC2 only test: nearly the same performances as the local SOLiD cluster (+10%) For only $3,2/hour, EC2 has nearly the same perf. as the local SOLiD cluster (16 cores, for 2H30)

44. UC 2: Acceleration of Financial Valuations 44

45. 45 A High Performance Solution  A Collaboration between Pricing Partners and ActiveEon  Price-it® Excel Accelerated by ProActive Parallel Suite®  A Global Solution: fully integrated with the same functionalities and interface as Price-it Excel while increasing its computing power  High Quality Service: from both companies

46. 46 How Does it Work? Price-it Computing Distribution

47. 47 Accelerated Price-it Performances Use Case: Bermuda Vanilla, Model American MC Test conditions:  One computation is split in 130 tasks that are distributed  Each task uses 300ko SequentialDistributed More than 3 times faster with only 4 nodes! 4 nodes5 nodes6 nodes7 nodes8 nodes9 nodes Even 6 times faster with 9 nodes!  Increased Productivity: Reduces Price-it Execution Time by 6 or more!

48. 48 Use Case 3: OMD2 48

49. 49 OMD2 Open Source Interfaces For Distribued Multi-Disciplinary Optimisations

50. 50 OMD2 : Open Source Interfaces For Distribued Multi-Disciplinaires Optimisations

51. 51 Distributed Workflow Scheduler WS BD PO Acquisition des maillages, etc Stratégie de résolution Boucle optimisation Calculs f(x) MailleurSolveur Soumission du PO Catia Visualisation stats, etc N1 N2 Ni1 Nik Ressources

52. 52 Les cas tests 3D Air Conditionning 2D Air Conditionning Cylinder HeadExternal Aerodynamic <1min CPU 10min CPU 100h CPU 1000h CPU

53. 53 UC 4. Parallel Scilab & Matlab Simulations 53

54. 54 Seamless Parallel & Distributed Scilab Dedicated resources LSF Static Policy Amazon EC2 EC2 Dynamic Workload Policy Desktops Timing Policy 12/24

55. 55 Interface ProActive  Scilab

56. 56 Interface ProActive  Matlab

57. 57 Conclusion: ProActive PACA GRID: Cloud Portal with GPUs in production

59. 59 The ProActive PACA Grid Platform (4) Total:  816 Cores  480 CUDA Cores  14.8TB Storage Publically Available Today

60. 60 GPU Computing Portal for Heterogeneous Resources Desktop Server Cluster Cloud

61. 61 New EU project: TEFIS 61

62. 62 New EU project: TEFIS Partners 62

63. 63 Interface to other TestBeds 63

64. 64 Interface to other TestBeds 64

65. 65 Conclusion 65

66. 66 Conclusion Free Professional Open Source Software Free Professional Open Source Software  EU Testbed Project: TEFIS  Regional Funding: PACA Lander  Portability: Windows, Linux, Mac  Versatility: Desktops, Grids, Clouds