1. Cloud et Accélération des
Applications Java avec
ProActive Parallel Suite
D. Caromel, et al.
Agenda
1. Background: INRIA, ActiveEon
2. ProActive OS Toolkit:
Programming, Scheduling, Resourcing
3. UC 1: Genomics+Cloud
4. UC 2: Finance
5. UC 3: IT, SOA Web Server
Log Analysis
Parallelism+Distribution+Virtualization
Enterprise Grids & Clouds

2. 1. Background
1. Background
2
2

3. OASIS Team & INRIA
 A joint team, Now about 35 persons
 2004: First ProActive User Group
 2009, April: ProActive 4.1, Distributed & Parallel:
From Multi-cores to Enterprise GRIDs
3
3

4. OASIS Team Composition (35)
 Researchers (5):  PostDoc (1):
 D. Caromel (UNSA, Det. INRIA)  Regis Gascon (INRIA)
 E. Madelaine (INRIA)  Engineers (10):
 F. Baude (UNSA)  Elaine Isnard (AGOS)
 F. Huet (UNSA)  Fabien Viale (ANR OMD2, Renault )
 L. Henrio (CNRS)  Franca Perrina (AGOS)
 PhDs (11): 

Germain Sigety (INRIA)
Yu Feng (ETSI, FP6 EchoGrid)
 Antonio Cansado (INRIA, Conicyt)  Bastien Sauvan (ADT Galaxy)
 Brian Amedro (SCS-Agos)  Florin-Alexandru.Bratu (INRIA CPER)
 Cristian Ruz (INRIA, Conicyt)  Igor Smirnov (Microsoft)
 Elton Mathias (INRIA-Cordi)  Fabrice Fontenoy (AGOS)
 Imen Filali (SCS-Agos / FP7 SOA4All)  Open position (Thales)
 Marcela Rivera (INRIA, Conicyt)
 Trainee (2):
 Muhammad Khan (STIC-Asia)
 Etienne Vallette d’Osia (Master 2 ISI)
 Paul Naoumenko (INRIA/Région PACA)
 Laurent Vanni (Master 2 ISI)
 Viet Dung Doan (FP6 Bionets)
 Virginie Contes (SOA4ALL)  Assistants (2):
 Guilherme Pezzi (AGOS, CIFRE SCP)  Patricia Maleyran (INRIA)
Located  Sophia Antipolis,(I3S)
in Sandra Devauchelle between
 + Visitors + Interns Nice and Cannes,
Visitors and Students Welcome! 4
4

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

6. 2. ProActive Parallel Suite
Dealing with Multi-Cores & Virtualization
6

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

8. ProActive Parallel Suite
 Three fully compatible modules:
Programming Scheduling
Resource Management
8

9. ProActive Programming:
Active Objects
99

10. ProActive : Active objects
A ag = newActive (“A”, […], VirtualNode)
V v1 = ag.foo (param);
V v2 = ag.bar (param);
...
v1.bar(); //Wait-By-Necessity
JVM JVM
A
v2 v1 ag
A
WBN!
V
Wait-By-Necessity
Java Object Active Object Req. Queue is a
Dataflow
Future Object Proxy Request Thread Synchronization
10
10 10

11. TYPED
ASYNCHRONOUS GROUPS
11
11

12. Broadcast and Scatter
Broadcast is the default behavior
Use a group as parameter, Scattered depends on rankings
cg ag
JVM
s
c1 c3
c3
c1 c2
c2
c1 c3
c2
JVM
JVM
ag.bar(cg); // broadcast cg
ProActive.setScatterGroup(cg);
ag.bar(cg); // scatter cg
JVM
12
12 12

13. Dynamic Dispatch Group
cg ag Slowest
c0 c2 c4 c6 c8 JVM
c1 c3 c5 c7 c9
Fastest
JVM
c0 c2 c4 c6 c8
c1 c3 c5 c7 c9
JVM ag.bar(cg);
JVM
13
13 13

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

15. ProActive Parallel Suite
 Workflows in Java
 Master/Workers
 SPMD
 Components
 …
15
15

16. GCM Fractal Deployment Standard
Interoperability:
 Protocols: Cloud will start with existing IT infrastructure,
 Rsh, ssh Build Non Intrusive Cloud with ProActive
 Oarsh, Gsissh
 Scheduler, and Grids:
 GroupSSH, GroupRSH, GroupOARSH
 ARC (NorduGrid), CGSP China Grid, EEGE gLITE,
 Fura/InnerGrid (GridSystem Inc.)
 GLOBUS, GridBus
 IBM Load Leveler, LSF, Microsoft CCS (Windows HPC Server 2008)
 Sun Grid Engine, OAR, PBS / Torque, PRUN
 Clouds:
 Amazon EC2
Denis Caromel
16 16

17. GCM Official Standardization
Grid Component Model
Overall, the standardization is supported by
industrials:
BT, FT-Orange, Nokia-Siemens, NEC,
Telefonica, Alcatel-Lucent, Huawei …
17
17

18. GCM Standardization
Fractal Based Grid Component Model
4 Standards:
1. GCM Interoperability Deployment
2. GCM Application Description
3. GCM Fractal ADL
4. GCM Fractal Management API
Denis Caromel
18 18

19. IC2D: Optimizing
19
19

20. IC2D
20
20

21. IC2D
21
21

22. ChartIt
22
22

23. Pies for Analysis and Optimization
23
23

24. Video 1:
IC2D Optimizing
Monitoring, Debugging, Optimizing
24

25. Scheduling & Resourcing
25
25

26. ProActive Scheduling
26
26 26

27. ProActive Scheduling Big Picture
 Multi-platform Graphical Client (RCP)
 File-based or LDAP authentication
ProActive  Static Workflow Job Scheduling, Native and
Scheduler Java tasks, Retry on Error, Priority Policy,
Configuration Scripts,…
ProActive
 Dynamic and Static node sources, Resource
Resource Manager
Selection by script, Monitoring and Control
GUI,…
 ProActive Deployment capabilities:
Desktops, Clusters, Clouds,…
RESOURCES
27

28. Scheduler: User Interface
28
28

29. Another Example : Picture Denoising
Split
Denoise Denoise Denoise Denoise
Merge
Job
•with selection on native executable availability (ImageMagik, GREYstoration)
• Multi-platform selection and command generation
•with file transfer in pre/post scripts
29

30. ProActive Resourcing
30
30 30

31. RESOURCING User Interface
31
31

32. Easily manage actual VMs
 Most of VMs can be dynamically provisioned:
 Start
 Stop
 Clone
 Destroy
 Supported VMs today:
 VMware,
 KVM
 Xen, Xen Server
 QMU
 Microsoft Hyper-V
 Provision Application containers:
 Manage applications with pre-defined VMs
Denis Caromel
32 32

33. Clusters to Grids to Clouds:
e.g. on Amazon EC2
33
33

34. Node source Usecase :
Configuration for external cloud with EC2
ProActive
Scheduler
ProActive
Resource Manager
Static Policy Timing Policy Dynamic
12/24 Workload Policy
LSF
Desktops EC2
Dedicated resources
Desktops Amazon EC2
34

35. Video 2:
Scheduler, Resource Manager
35

36. ProActive Parallel Suite
Three fully compatible modules
Programming Scheduling
Resourcing
Scheduling
Resource Management
Clutch Power:
Solid Building Blocks
for Flexible Solutions
36

37. Use Cases
37
37

38. Use Case 1:
Genomics
38

39. Resources set up
SOLID
machine from
16
nodes
Cluster
Desktops
Nodes
can be
dynamically
added!
EC2 Clouds
39
39

40. 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
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
On going the local SOLiD cluster (16 cores, for 2H30)
Benchmarks on Windows Desktops and HPCS 2008 …
40

41. Benchmark: local vs. hybrid cloud
Use case: 3 runs performed in parallel containing a total of 28,5
millions of reads to be matched against the human genome
 SOLID nodes only  SOLID and EC2 nodes
Standard configuration 12 SOLiD nodes
using SOLID embedded 12 EC2 machines
nodes: 12 (type: “mlarge”, 2 nodes each)
 Total computation time:  Total computation time:
12.5 hours 8 hours
Gain: 4,5 hours (36% faster)
EC2 costs: $40
41

42. Benchmark: local vs. EC2 cloud
Execution time Cost
(min) (US$)
Standard PBS config 300 NA
ProActive Amazon EC2 340 20 US$
For only $3,2/hour, the EC2 setup has nearly the
same performances as the local SOLiD cluster
42

43. UC 2: Acceleration of
Financial Valuations
43

44. 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
44

45. Some Technical Facts
 Price-It®
 C++ library developed by Pricing Partners
 Pricing solution dedicated to highly complex
financial derivatives
 Specification and Constraints
 Accelerate Price-It® Excel product
 Built on Price-It® library, this product integrates
an interface with Excel for input data management
and results display
 Focus on highly parallelizable Greek computation
 Operating system: Windows
45

46. How Does it Work?
Price-it Computing Distribution
Price-it Price-it Regular Price-it Excel
Excel Excel
Interface
ProActive Automatic execution
Scheduler via job scheduler
Pool of shared
resources
46

47. Accelerated Price-it Performances
 Increased Productivity: Reduces Price-it Execution
Time by 6 or more!
Use Case: Bermuda
Vanilla, Model
American MC
More than 3 times faster
with only 4 nodes! Test conditions:
 One computation
is split in 130
tasks that are
Even 6 times faster distributed
with 9 nodes!  Each task uses
300ko
4 nodes 5 nodes 6 nodes 7 nodes 8 nodes 9 nodes
Sequential Distributed
47

48. UC 3: SOA Analysis
of Web Server Logs
48

49. Parallel Services
 Separation: BPEL – Parallel Serv. – Task Flow
 Standards et Portable
 Flexibility High level Business Process
Domain specific Service
Other
… Basic Service
Job
Scheduling Parameter
Divide & Other …
Conquer Operational
Sweeping
Service Operational Services
…
Parallel Services
Scheduling `of Taskflow Jobs
Scheduling
Parameter
Sweeping
Service
Resource Man.
Enterprise Grid
49
49

50. Cas d’études AMADEUS : Démonstration
BPEL Scheduler
Log Parsing
Log Parsing
Log Parsing
RM Grille Log Parsing
Log Parsing
WS DB
Log Transfer Application ID Log Transfer
Application
ID Logs stockés
Log Parsing Log Parsing
Agos Web
Client
Admin Admin WS Scheduler/ MySQL
BPEL Console RM Console
Console Interface
BPEL/WS
Web Client 50
50

51. Grid Monitoring integration
Job Business
X Availability Center
Jobs
Scheduler Scheduler
Job DB
Y
Discovery Universal
JDBC
Resource DDM Create CMDB
interface
Manager Configuration
Discover Jobs,
Management
Tasks &
System
Resources
JMX interface information
Collect Grid Jobs
Type status
indicators
Indicators
Collect Grid SiS
Components
NATIVE
VM1 VM2
1
VM3 statistics
indicators
Pool of nodes in the grid
NATIVE NATIVE
VM4 VMx
3 2
51

52. AGOS Platform Management
HP- Business Availability Center
Tasks scheduler &
(HP-BAC) Resources manager
• Monitoring of the entire platform • Integration with grid
• Cover all layers in the scope components
•Provide monitoring dashboard and • Grid insights through indicator
reports collection and running jobs on
grid resources
52

53. Conclusion
53
53

54. ProActive in Cloud Stack
IaaS
54

55. Conclusion
Flexibility
Clutch Power
Portability:
Windows, Linux, Mac
Versatility:
Java Parallel Multi-Platform Resource
Desktops, Grids, Clouds
Toolkit Job Scheduler Manager
Free Professional
Open Source Software
Multi-Core: Ready for the next revolution
Virtualization: Dynamic Load Balancing onto VMs
Cloud: Smooth transition available (Desktop , Server, Cluster)
Dynamic Workload Management with Virtualization
55

56. 56
56

57. 4. Cloud Seeding
57

58. Cloud Seeding with ProActive
 Amazon EC2 Execution
 Cloud Seeding strategy to mix heterogeneous
computing resources :
 External GPU resources
58

59. Cloud Seeding with ProActive
CPU nodes
Web Interface ProActive Scheduler
+ Resource Manager
Amazon EC2
User
Noised video file
GPU nodes
59

60. Cloud Seeding with ProActive
CPU nodes
Web Interface ProActive Scheduler
+ Resource Manager
Amazon EC2
User
User submit its noised video to the web
interface GPU nodes
60

61. Cloud Seeding with ProActive
CPU nodes
Web Interface ProActive Scheduler
+ Resource Manager
Amazon EC2
User
Web Server submit a denoising job the
ProActive Scheduler GPU nodes
61

62. Cloud Seeding with ProActive
CPU nodes
Web Interface ProActive Scheduler
+ Resource Manager
Amazon EC2
User
CPU nodes are used to split the video into
smaller ones GPU nodes
62

63. Cloud Seeding with ProActive
CPU nodes
Web Interface ProActive Scheduler
+ Resource Manager
Amazon EC2
User
CPU nodes are used to split the video into
smaller ones GPU nodes
63

64. Cloud Seeding with ProActive
CPU nodes
Web Interface ProActive Scheduler
+ Resource Manager
Amazon EC2
User
GPU nodes are responsible to denoise
these small videos GPU nodes
64

65. Cloud Seeding with ProActive
CPU nodes
Web Interface ProActive Scheduler
+ Resource Manager
Amazon EC2
User
GPU nodes are responsible to denoise
these small videos GPU nodes
65

66. Cloud Seeding with ProActive
CPU nodes
Web Interface ProActive Scheduler
+ Resource Manager
Amazon EC2
User
CPU nodes merge the denoised video parts
GPU nodes
66

67. Cloud Seeding with ProActive
CPU nodes
Web Interface ProActive Scheduler
+ Resource Manager
Amazon EC2
User
CPU nodes merge the denoised video parts
GPU nodes
67

68. Cloud Seeding with ProActive
CPU nodes
Web Interface ProActive Scheduler
+ Resource Manager
Amazon EC2
User
The final denoised video is sent back to
the user GPU nodes
68

69. 6. SOA, SLA and QoS
69
69 69

70. Standard system at Runtime: No Sharing
NoC: Network On Chip
Proofs of Determinism
70
70 70

71. AGOS: Grid Architecture for SOA
Building a Platform for Agile SOA with Grid
 AGOS Solutions
In Open Source with Professional Support
71
71

72. AGOS Infrastructure Management
HP Systems Insight Manager (HP- Citrix XenCenter
SIM)
• Monitoring of entire infrastructure • Hypervisor and VM
• Communicates with upper layer management
management software (HP BAC) • Communicates with upper
layer management software
(HP BAC)
72

73. AGOS and HP Management tools Integration
Services
Processes and Jobs
Grid Components
Monitoring
Discovery
Scheduler, Resource Manager
Hypervisors/Virtual Machines
Xen and Vmware Hosts and Guests
Hardware Infrastructure
Servers, Storage, Network Components
73