The document summarizes a PhD thesis defense presentation on soCloud, a distributed multi-cloud platform. The presentation covers: 1) The soCloud model, which extends the Service Component Architecture (SCA) model to support annotations for non-functional requirements to deploy components as execution units across multiple clouds. 2) The soCloud platform, which is a distributed component-based PaaS that manages portability, provisioning, elasticity, and high availability across clouds using a reactive architecture. 3) Validation of the soCloud model and platform through modeling applications and providing runtime support to manage applications in multi-cloud environments.

1. soCloud: distributed multi-cloud
platform for deploying, executing and
managing distributed applications
Fawaz PARAISO
PhD Defense
Advisors: Lionel Seinturier, Philippe Merle
University Lille 1, Inria, SPIRALS research team

2. 2
Cloud computing in nutshell
Virtualization
Pay-per-use
On-demand
Elasticity

3. 3
Context and motivation
Application fil rouge
Go to the Cloud !
View%
Compute%
Storage%
Developer

4. 4
So many problems !
Vendor Lock-in
Failures
Geo-location
Cloud-specific services

5. 5
solution: Multi-Cloud
Multi-Cloud
Do not put all your eggs in one basket
Why not applying this precept of caution for cloud
computing?

6. 6
Definition
• Multi-Cloud Computing
✤ using multiple cloud providers
✤ independent
✤ no agreement between providers
What is Multi-Cloud ?

7. 7
Multi-CloudContext and motivation
74%Enterprises have a Multi-Cloud strategies

8. 8
Context and motivation
Multi-cloud is supposed to be the solution but…
Towards Multi-Cloud Computing

9. Outline
9
1.Context and motivation
2.Challenges
3.State of the art
4.Contributions
4.1.soCloud Model
4.2.soCloud Platform
5.Validation
6.Conclusion & Perspectives

10. 10
Multi-cloud Provisioning
Multi-cloud Elasticity
Multi-cloud High-Availability
Multi-cloud Portability
Challenges

11. 11
Multi-cloud Portability
Multi-cloud
Provisioning
Multi-cloud Elasticity
Multi-cloud High
availability
AppApp
App App
Your application/data
App
Write once, deploy anywhere
without any modification
Challenges

12. 12
Multi-cloud Portability
Multi-cloud
Provisioning
Multi-cloud Elasticity
Multi-cloud High
availability
Resources
App Deployment
Challenges

13. 13
Multi-cloud Portability
Multi-cloud
Provisioning
Multi-cloud Elasticity
Multi-cloud High
availability
Resources
Provisioning
Provisioning
Provisioning
Challenges

14. 14
Multi-cloud Portability
Multi-cloud
Provisioning
Multi-cloud Elasticity
Multi-cloud High
availability
App deployment
Challenges

15. 15
Multi-cloud Portability
Multi-cloud
Provisioning
Multi-cloud Elasticity
Multi-cloud High
availability
Challenges

16. 16
Challenges
Multi-cloud Portability
Multi-cloud
Provisioning
Multi-cloud Elasticity
Multi-cloud High
availability
HA

17. 17
Multi-cloud Portability
Multi-cloud
Provisioning
Multi-cloud Elasticity
Multi-cloud High
availability
HA
HA
HA
Challenges

18. Outline
18
1.Context and motivation
2.Challenges
3.State of the art
4.Contributions
4.1.soCloud Model
4.2.soCloud Platform
5.Validation
6.Conclusion Perspectives

19. 19
Approach Layer
Multi-Cloud
Portability
Multi-Cloud
Provisioning
Multi-Cloud
Elasticity
Multi-Cloud
High-Availability
moSAIC PaaS + + + -
STRATOS IaaS - + + -
MODAClouds PaaS + - - +
CompatibleOne IaaS + + - +
Cloud4SOA PaaS + + - -
soCloud PaaS + + + +
State of the art

20. 20
State of the art
Approach Layer
Multi-Cloud
Portability
Multi-Cloud
Provisioning
Multi-Cloud
Elasticity
Multi-Cloud
High-Availability
moSAIC PaaS + + + -
STRATOS IaaS - + + -
MODAClouds PaaS + - - +
CompatibleOne IaaS + + - +
Cloud4SOA PaaS + + - -
soCloud PaaS + + + +

21. 21
State of the art
Approach Layer
Multi-Cloud
Portability
Multi-Cloud
Provisioning
Multi-Cloud
Elasticity
Multi-Cloud
High-Availability
moSAIC PaaS + + + -
STRATOS IaaS - + + -
MODAClouds PaaS + - - +
CompatibleOne IaaS + + - +
Cloud4SOA PaaS + + - -
soCloud PaaS + + + + +

22. 22
State of the art
Approach Layer
Multi-Cloud
Portability
Multi-Cloud
Provisioning
Multi-Cloud
Elasticity
Multi-Cloud
High-Availability
moSAIC PaaS + + + -
STRATOS IaaS - + + -
MODAClouds PaaS + - - +
CompatibleOne IaaS + + - +
Cloud4SOA PaaS + + - -
soCloud PaaS ++ ++ ++ ++

23. Outline
23
1.Context and motivation
2.Challenges
3.State of the art
4.Contributions
4.1.soCloud Model
4.2.soCloud Platform
5.Validation
6.Conclusion Perspectives

24. 24
soCloud
Platform
EC2
Azure
Heroku
Developer
soCloud Overview
soCloud Model
distributed applications

25. Outline
25
1.Context and motivation
2.Challenges
3.State of the art
4.Contributions
4.1.soCloud Model
4.2.soCloud Platform
5.Validation
6.Conclusion Perspectives

26. 26
Objectives
‘’Provides a model to design a distributed
applications in a simple and concise
manner for a Multi-Cloud environment’’

27. 27
Features
Identify requirements for engineering distributed application for the Multi-Cloud
environments
Multi-Cloud
Portability
Abstraction
Standard
Structure
Multi-Cloud
Provisioning
Placement
Resources
Granularity
Multi-Cloud
Elasticity
DSL
Simple
Multi-Cloud
High-availability
Failures
Diversity

28. 28
soCloud Model
Extended SCA Model

29. 29
SCA
• Service Component Architecture (SCA)
✤ Set of OASIS specifications
✤ Distributed applications
✤ Using SOA
ZIP File
Composite Component
Service Reference
Property Wire
Legend
View
Compute
Storage
SCA Contribution

30. 30
soCloud Model based on SCA
Component(
Property(
Implementa/on(
Reference(
PolicyIntent(
PolicySet(
Opera/on(
Interface(
Interface.java(
Binding(
CallBack(
Wire(
ConstrainingType(
Implementa/on.Contribu/on(
+source(
+target(
+name(
+autowire(
+uri(
+name(
+name(
+requires(
+policySets(
+name(
+mul5plicity(
+wireByImpl(
+conversa5onal( +callBackInterface(
+interface(
Interface.WSDL(
+callBackInterface(
+interface(
+requires(
+policySets(
+name(
+name(
+mustSupply(
+many(
+value(
+element(
0..*(
0..*(
0..*(
0..*(
0..*(
0..*(
0..*(
0..*(
0..*(
1..1(
0..*(
0..*( 0..*(
0..*(
0..*(
0..*(
1..*(
1..*(
1..*(
0..1(
1..*(
0..*(
0..*(
requires(
requires(
requires(
requires(
target(
policySets(
policySets(
policySets(
+requires(
+policySets(
Composite(
+name(
+autowire(
+targetNamespace(
+local(
0..1(
0..1(
Service(
+name(
+name(
+value(
Annota/on( Implementa/on.Composite(
+name(
Implementa/on.BPEL(
+process(
Implementa/on.C++(
+class(
Implementa/on.Java(
+class(
0..*(
+name
+value
Annotaon'
Implementa)on.Contribu)on1
+name
Extension 2
Extension 1

31. 31
soCloud Model based on SCA
+name
+value
Annotaon'
Extension 2
Implementa)on.Contribu)on1
+name
Extension 1
Why extend SCA model ?

32. 32
soCloud Model: implementation
Implementa)on.Contribu)on1
+name
• Provides high level conceptual view to
a component
• Allows the deployment of the
component as execution unit
component
• Structured components of distributed
applications
Extension 1

33. 33
soCloud Model: annotations
• The SCA model does not allow us to
take into account these non-functional
requirements
• Allowing to associate non-functional
requirements to a component
component
@
+name
+value
Annotaon'
Extension 2

34. 34
soCloud Model: annotations
Annotation
+name
+value
Placement
annotation
Execution
annotation
Availability
annotation
Elasticity
annotation
location closer vm database replication elasticity
service

35. 35
soCloud Model: annotations
Annotation
+name
+value
Placement
annotation
Execution
annotation
Availability
annotation
Elasticity
annotation
location closer vm database replication elasticity
service
location closer

36. 36
soCloud Model: annotations
Placement
annotation
location closer
1@location = ‘value’
2 @closer = ‘value’

37. 37
soCloud Model: annotations
Any
AmericaAfrica Asia Europe Oceania
Clouds provider
Amazon EC2 Windows Azure
New york California FranceIrlandAmazon IrlandSingapourSouth Africa Australia
Paris Roubaix
(1)
(2)
(3)
1@location = ‘value’

38. 38
soCloud Model: annotations
C1 C2
Latency
@closer =‘C2’
2 @closer = ‘value’

39. 39
soCloud Model: annotations
Annotation
+name
+value
Placement
annotation
Execution
annotation
Availability
annotation
Elasticity
annotation
location closer vm database replication elasticity
service
vm database
service

40. 40
soCloud Model: annotations
1 @vm = ‘type_vm’
2 @database = ‘name - version’
3 @service = ‘name - version’
Execution
annotation
vm database
service
1 @vm = ‘micro’
2 @database = ‘MySQL’
3 @service = ‘IronMQ - 2.8.9’
Example
Optional

41. 41
soCloud Model: annotations
Annotation
+name
+value
Placement
annotation
Execution
annotation
Availability
annotation
Elasticity
annotation
location closer vm database replication elasticity
service
replication

42. 42
soCloud Model: annotations
1 @replication = ‘number’
Availability
annotation
replication
C
@replication=5
become
LB
C
Example

43. 43
soCloud Model: annotations
Annotation
+name
+value
Placement
annotation
Execution
annotation
Availability
annotation
Elasticity
annotation
location closer vm database replication elasticity
service
elasticity

44. 44
soCloud Model: annotations
Elasticity
annotation
elasticity
1 @elasticity = ‘description’
C
C
scale outscale in
A DSL for describing elasticity

45. 45
soCloud Model: elasticity language
Event Action Condition
scaling up when (
average (cpuUsage,120s) 80%
)
minimize availability when (
totalCost(costCompute,24 h) 900 )
)
Elasticity is expressed on the Resources, Cost, Quality

46. 46
soCloud Model: elasticity language
Trigger
scaling in 5 at ( 20:00 Friday)

47. 47
soCloud Model: annotations
3-Tiers Application
@elasticity=‘’Scaling in
When ( totalCost(computeCost, 24 h) 200 )’’
View
Compute
Storage
@closer=‘’Storage ‘’
@vm=‘’xlarge-Ubuntu‘’
@location=‘’France‘’
@elasticity=‘’Scaling in
When ( totalCost(computeCost, 24 h) 900 )’’
View
Compute
@closer=‘’Storage ‘’
@vm=‘’xlarge-Ubuntu‘’
Storage
@database=‘’MySQL’’
@location=‘’France’’

48. 48
Summary
• We show how we use annotation to describe non-functional
properties and manage each component as unit of execution
• New language is proposed to effectively express the elasticity
• Paraiso F, Merle P and Seinturier L : soCloud : A service- oriented component-based
PaaS for managing portability, provisioning, elasticity et high availability across
multiple clouds. Springer Computing Journal (Submitted)!
• Haderer N, Paraiso F, Ribeiro C, Merle P, Rouvoy R and Seinturier L : A Cloud-based
Infrastructure for Crowd-sourcing Data from Mobile Devices. Springer Book (To
appear)

49. Outline
49
1.Context and motivation
2.Challenges
3.State of the art
4.Contributions
4.1.soCloud Model
4.2.soCloud Platform
5.Validation
6.Conclusion Perspectives

50. 50
soCloud Platform
• The expectations in term of execution support for
distributed applications built with soCloud Model are
differents
• We need to provide a Platform that manages:
✤ Multi-Cloud environments
✤ Distributed applications in Multi-Cloud environments

51. 51
soCloud Platform: concept
We need to build Multi-Cloud Platform that:
• react to load — Scalable
• react to event — Event-Driven
• react to failure — Fault-Tolerance
• react to change — Responsive
Reactive, flexible and self management platform
• self management — Autonomic
• is flexible — Component-based

52. 52
soCloud Platform
• soCloud Platform is a distributed component-based PaaS
for managing
✤ Portability
✤ Provisioning
✤ Elasticity
✤ High-availability
Components

53. 53
soCloud platform high level view
soCloud
master
soCloud agent
User …
soCloud applications

54. 54
soCloud Platform: Multi-Cloud centric Architecture
Trend in the soCloud Platform Architecture
single Cloud centric Architecture
soCloud agent
soCloud master soCloud master
soCloud agent
Multi-Cloud centric Architecture
soCloud agent soCloud agent
soCloud master

55. 55
soCloud Platform detail view
Monitoring
soCloud master
soCloud agent
LoadBalancer
Service
Deployer
Node Provisioning
PaaS Deployment
SaaS Deployment
Controller
Workload
Manager
Constraint
Validator

56. 56
soCloud Platform: Fault Tolerance
Application level
Replication in
different clouds
Platform level
Replication in
different clouds
Let it Crash

57. 57
soCloud Platform: Fault Tolerance
To achieve this
1. Transparency is the ultimate goal [Waldo et. al]
2. Automatic component and applications replication [Waldo et. al]
3. All replications are equal and deterministic [Waldo et. al]
[Waldo et. al]-Classic paper: A Note On Distributed Computing

58. 58
soCloud Platform: Replication features
• A cluster of N servers distributed across
several Clouds
• Any (exactly one) component can be leader
• Active replication by the leader
• Consensus election of the leader
• Automatic failover
• Automatic recovery

59. 59
soCloud Platform: deployment stack
IaaS
soCloud
master
agent
SCA container
Servlet container
Java runtime
Resources
Linux/OS

60. 60
soCloud Platform: deployment stack
soCloud
master
agent
SCA container
Servlet container
PaaS

61. 61
Summary
• Runtime support for managing Multi-Cloud portability,
provisioning, elasticity and high-availability
• Reactive Platform
• PARAISO Fawaz et.al : A federated multi-cloud PaaS infrastructure. In IEEE 5th International
Conference on Cloud Computing (CLOUD), pages 392–399., Hawaii IEEE, 2012.
• PARAISO Fawaz et al.: Managing elasticity across multiple cloud providers. In Proceedings of the
2013 international workshop on Multi-cloud applications and federated clouds, pages 53–60.
ACM, 2013.

62. 62
Outline
1.Context and motivation
2.Challenges
3.State of the art
4.Contributions
4.1.soCloud Model
4.2.soCloud Platform
5.Validation
6.Conclusion Perspectives

63. 63
Validation
soCloud Model
soCloud Platform

64. 64
Validation: soCloud Model
1. APISENSE application
Modeling of three concrete applications using the
soCloud Model
2. DiCEPE application
3. P2P Monitoring application

65. 65
Validation: soCloud Model
Sensing'Storage'
APISENSE'
Central(Node(
Sensing'Node'Sensing'Node'
store data
Publishtask
Publishtask
Subscribe taskSubscribe task
Create task
Downloadsensingtask
Reward
Send data
Senddata
Send data Send data
1. APISENSE application [Nicolas Haderer]

66. 66
Validation: soCloud Model
• Geo-location
✤ Paris
• Unpredictable growth of smartphones
• Availability despite failures
• Cost control

67. 67
Validation: soCloud Model
composite name=Application-APISENSE
component name=SensingNode
implementation.contribution contribution=sensingnode.zip/
reference name=compute target=CentralNode/compute/
reference name=storage target=SensingStorage/storage/
annotation name=locationParis/annotation
annotation name=replication2/annotation
annotation name=elasticity
scaling in when (totalCost(computeCost, 24h) 1000)
/annotation
/component
/composite

68. 68
Summary
• The soCloud Model has enabled us to build an App for collecting
data from smartphones, an App to integrate heterogenous
CEP Engines and make Big Data, and finally a P2P distributed
App
• PARAISO Fawaz et.al : A federated multi-cloud PaaS infrastructure. In IEEE 5th International
Conference on Cloud Computing (CLOUD), pages 392–399., Hawaii IEEE, 2012.
• PARAISO Fawaz et. al.: A Middleware Platform to Federate Complex Event Processing. In Sixteenth
IEEE International EDOC Conference, pages 113–122, Beijing, China, septembre 2012.
Springer.

69. 69
Validation: soCloud Platform
1. Portability
2. High-availability
3. Elasticity
4. Overhead introduced by soCloud

70. 70
Validation: soCloud Platform
Portability

71. 71
Validation: soCloud Platform
Deployed on
10 Clouds
IaaS and PaaS

72. 72
Validation: soCloud Platform
High-availability

73. 73
Validation: soCloud Platform
=
MTBF*
MTBF + MTTR**
Availability
MTTR** = Mean Time To Recover
MTBF* = Mean Time Between Failure
[Marcus et. al.]
[Marcus et. al.] : Blueprints for High availability

74. 74
Validation: soCloud Platform
MTTR* (Hour) MTTR (Minute) Ratio
soCloud 0.06 Hour 3.6 Minutes -
Public clouds [IWGCR] 7.5 Hours 450 Minutes 125
MTTR* = Mean Time To Recover
[IWGCR] = International Working Group on Cloud Computing Resiliency. http://iwgcr.org

75. 75
Validation: soCloud Platform
Availability
soCloud
Public clouds
=8760
8760 + 0.06
99.999%
=8760
8760 + 7.5
99.914%
If it is assumed that a failure occurs once per year
MTBF = 8760 Hours

76. 76
Validation: soCloud Platform
Elasticity

77. 77
Flash crowd effect
0
10000
20000
30000
40000
50000
60000
1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171 181 191
Time%(slotof10seconds)%
Number%of%requests%
(a)
3-Tiers application was deployed on ten cloud providers
Total Number of Request = 3020000
Phase 1 Phase 2

78. 78
Flash crowd effect without soCloud elasticity
0
500
1000
1500
2000
2500
3000
3500
4000
4500
0 20 40 60 80 100 120 140 160 180 200
Number'of'failed'requests'
Time'(slotof10seconds)'
1.3% of requests are failed that correspond to 34039
Response Time = 65.90 s

79. 79
Flash crowd effect with soCloud elasticity
0
10
20
30
40
50
60
70
1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171 181 191
Response'(me'(s)'
Time'(slotof10seconds)'
Applica(on'replica(on'
Balance'load'
Response Time = 37.3 s Response Time = 23.38 s
Phase 1 Phase 2
Without soCloud elasticity, the Response Time = 65.90 s
No request has failed

80. 80
Overhead introduced by soCloud
Implementation Execution time Overhead introduced by soCloud
(Application + FraSCAti) 10.85 sec -
(Application + FraSCAti +
soCloud)
11,10 sec 2.3%
APP
APP

81. 81
Overhead introduced by soCloud
Implementation Execution time Overhead introduced by soCloud
(Application + FraSCati) 10.85 sec -
(Application + FraSCati +
soCloud)
11,10 sec 2.3%
The benefit provided by the soCloud Platform outweighs the
difference in the execution time

82. 82
Summary
• Reactivity face:
✤ Failures (High-availability)
✤ Flash crowd effect (Elasticity)
• Negligible Overhead introduced.

83. Outline
83
1.Context and motivation
2.Challenges
3.State of the art
4.Contributions
4.1.soCloud Model
4.2.soCloud Platform
5.Validation
6.Conclusion Perspectives

84. 84
Conclusion
soCloud Model
• We use annotations to express non-functional
requirements.
• New language is proposed to effectively express the
elasticity.
• The soCloud Model is illustrated on three distributed
applications deployed in Multi-Cloud environments.
• Multi-Cloud PaaS for deploying, executing and
managing distributed application.
• It was deployed on ten IaaS/PaaS clouds providers.
• soCloud Platform is capable of providing Multi-Cloud
high-availability and elasticity to applications
deployed on it.
soCloud Platform

85. 85
Perspectives
Short-term further work
• The high-availability management despite software
bugs.
• The elasticity management using reinforcement
learning.
• Security for Multi-Cloud.
• Sharing state between replicates.
• Take into account changes of the underlying platforms.
Further Research Directions

86. 86
Thank you !!!
• PARAISO Fawaz, HADERER Nicolas, MERLE Philippe, ROUVOY Romain and SEINTURIER
Lionel : A federated multi-cloud PaaS infrastructure. In IEEE 5th International Conference on
Cloud Computing (CLOUD), pages 392–399. IEEE, 2012.
• PARAISO Fawaz, HERMOSILLO Gabriel, ROUVOY Romain, MERLE Philippe, SEINTURIER
Lionel : A Middleware Platform to Federate Complex Event Processing. In Sixteenth IEEE
International EDOC Conference, pages 113–122, Beijing, China, septembre 2012. Springer.
• PARAISO Fawaz, MERLE Philippe and SEINTURIER Lionel : Managing elasticity across multiple
cloud providers. In Proceedings of the 2013 international workshop on Multi-cloud
applications and federated clouds, pages 53–60. ACM, 2013.
• PARAISO Fawaz, MERLE Philippe and SEINTURIER Lionel : soCloud : A service- oriented
component-based PaaS for managing portability, provisioning, elasticity et high availability across
multiple clouds. Springer Computing Journal (To appear)!
• HADERER Nicolas, PARAISO Fawaz, RIBEIRO Christophe, MERLE Philippe, ROUVOY
Romain and SEINTURIER Lionel : A Cloud-based Infrastructure for Crowd-sourcing Data
from Mobile Devices. Springer Review (To appear)