socloud: distributed multi-cloud platform for deploying ... · socloud: distributed multi-cloud...

soCloud: distributed multi-cloud platform for deploying, executing and

managing distributed applicationsFawaz PARAISO

PhD Defense

Advisors: Lionel Seinturier, Philippe Merle

University Lille 1, Inria, SPIRALS research team

�2

Cloud computing in nutshellVirtualization

Pay-per-use

On-demand

Elasticity

�3

Context and motivation

Application fil rouge

Go to the Cloud !

View%

Compute%

Storage%

Developer

�4

So many problems !

Vendor Lock-in

Failures

Geo-location

Cloud-specific services

�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

Definition

• Multi-Cloud Computing ✤ using multiple cloud providers ✤ independent ✤ no agreement between providers

What is Multi-Cloud ?

�7

Multi-CloudContext and motivation

74% Enterprises have a Multi-Cloud strategies

�8

Context and motivation

Multi-cloud is supposed to be the solution but…

Towards Multi-Cloud Computing

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

Multi-cloud Provisioning

Multi-cloud Elasticity

Multi-cloud High-Availability

Multi-cloud Portability

Challenges

��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


availability

Resources

App Deployment

Challenges

�13


availability

Resources

Provisioning

Provisioning

Provisioning

Challenges

�14


availability

App deployment

Challenges

�15


availability

Challenges

�16

ChallengesMulti-cloud Portability Multi-cloud

Provisioning Multi-cloud Elasticity Multi-cloud High availability

HA

�17


availability

HA

HA

HA

Challenges

Outline

�18


2.Challenges

3.State of the art

4.Contributions

4.1.soCloud Model


5.Validation


�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

State of the artApproach Layer Multi-Cloud

PortabilityMulti-Cloud

ProvisioningMulti-Cloud Elasticity


moSAIC PaaS + + + -





soCloud PaaS + + + +

�21





moSAIC PaaS + + + -





soCloud PaaS + + + + +

�22





moSAIC PaaS + + + -





soCloud PaaS ++ ++ ++ ++

Outline

�23


2.Challenges

3.State of the art

4.Contributions

4.1.soCloud Model


5.Validation


�24

soCloud"Platform

EC2

Azure

HerokuDeveloper

soCloud Overview

soCloud Model

distributed applications

Outline

�25


2.Challenges

3.State of the art

4.Contributions

4.1.soCloud Model


5.Validation


�26

Objectives

‘’Provides a model to design a distributed applications in a simple and concise

manner for a Multi-Cloud environment’’

�27

FeaturesIdentify 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

soCloud Model

Extended SCA Model

�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

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&

Annota&on'

Implementa)on.Contribu)on1+name&

Extension 2

Extension 1

�31

soCloud Model based on SCA

+name&+value&

Annota&on'

Extension 2

Implementa)on.Contribu)on1+name&

Extension 1

Why extend SCA model ?

�32

soCloud Model: implementationImplementa)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

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&

Annota&on'

Extension 2

�34

soCloud Model: annotationsAnnotation+name+value

Placement "annotation

Execution "annotation

Availability "annotation

Elasticity "annotation

location closer vm database replication elasticity

service

�35




Availability"annotation



service

location closer

�36

soCloud Model: annotationsPlacement "annotation

location closer

1@location = ‘value’"2 @closer = ‘value’

�37


Any

AmericaAfrica Asia Europe Oceania

Clouds providerAmazon EC2 Windows Azure

New york California FranceIrlandAmazon IrlandSingapourSouth Africa Australia

Paris Roubaix

(1)

(2)

(3)

1@location = ‘value’

�38


C1 C2

Latency

@closer =‘C2’

2 @closer = ‘value’

�39







service

vm database

service

�40


1 @vm = ‘type_vm’"2 @database = ‘name -> version’"3 @service = ‘name -> version’


vm database

service 1 @vm = ‘micro’"2 @database = ‘MySQL’"3 @service = ‘IronMQ -> 2.8.9’

Example

Optional

�41







service

replication

�42

soCloud Model: annotations1 @replication = ‘number’


replication

C

@replication=5

become

LB

C

Example

�43







service

elasticity

�44



elasticity

1 @elasticity = ‘description’

C

C

scale outscale in

A DSL for describing elasticity

�45

soCloud Model: elasticity languageEvent 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

soCloud Model: elasticity languageTrigger

scaling in 5 at ( 20:00 Friday)

�47

soCloud Model: annotations3-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

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)

Outline

�49


2.Challenges

3.State of the art

4.Contributions

4.1.soCloud Model


5.Validation


�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

soCloud Platform: conceptWe 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

soCloud Platform• soCloud Platform is a distributed component-based PaaS

for managing ✤ Portability ✤ Provisioning ✤ Elasticity ✤ High-availability

Components

�53

soCloud platform high level view

soCloud "master

soCloud agentUser …

soCloud applications

�54

soCloud Platform: Multi-Cloud centric ArchitectureTrend 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

soCloud Platform detail view

Monitoring

soCloud master

soCloud agent

Load

Bal

ance

r

Service "Deployer

Node Provisioning

PaaS Deployment

SaaS Deployment

Controller

Workload "Manager

Constraint "Validator

�56

soCloud Platform: Fault Tolerance

Application level

Replication in different clouds

Platform level

Replication in different clouds

Let it Crash

�57

soCloud Platform: Fault ToleranceTo 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

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

soCloud Platform: deployment stack

IaaS

soCloud

master

agent

SCA container

Servlet container

Java runtime

Resources

Linux/OS

�60

soCloud Platform: deployment stack

soCloud

master

agent

SCA container

Servlet containerPaaS

�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

Outline


2.Challenges

3.State of the art

4.Contributions

4.1.soCloud Model


5.Validation


�63

Validation

soCloud Model

soCloud Platform

�64

Validation: soCloud Model

1. APISENSE application

Modeling of three concrete applications using the soCloud Model

2. DiCEPE application

3. P2P Monitoring application

�65


Sensing'Storage'

APISENSE'Central(Node(

Sensing'Node'Sensing'Node'

store data

Publ

ish

task

Publ

ish

task

Subscribe taskSubscribe task

Create taskD

ownl

oad

sens

ing

task

Rew

ard

Send data Send

dat

a

Send data Send data

1. APISENSE application [Nicolas Haderer]

�66


• Geo-location ✤ Paris

• Unpredictable growth of smartphones

• Availability despite failures

• Cost control

�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="location">Paris</annotation> <annotation name="replication">2</annotation> <annotation name="elasticity"> scaling in when (totalCost(computeCost, 24h) > 1000) </annotation> </component> </composite>

�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

Validation: soCloud Platform

1. Portability 2. High-availability 3. Elasticity 4. Overhead introduced by soCloud

�70


Portability

�71


Deployed on

10 Clouds

IaaS and PaaS

�72


High-availability

�73


= MTBF*MTBF + MTTR**

Availability

MTTR** = Mean Time To RecoverMTBF* = Mean Time Between Failure

[Marcus et. al.]

[Marcus et. al.] : Blueprints for High availability

�74


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


Availability

soCloud

Public clouds

=87608760 + 0.06 99.999%

=87608760 + 7.5 99.914%

If it is assumed that a failure occurs once per year

MTBF = 8760 Hours

�76


Elasticity

�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%(slot"of"10"seconds)%

Num

ber%o

f%req

uests%

(a)"

3-Tiers application was deployed on ten cloud providers

Total Number of Request = 3020000

Phase 1 Phase 2

�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"

Num

ber'o

f'failed'requ

ests'

Time'(slot"of"10"seconds)'

1.3% of requests are failed that correspond to 34039

Response Time = 65.90 s

�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"

Respon

se'(me'(s)'

Time'(slot"of"10"seconds)'

Applica(on'replica(on'

Balance'load'

Response Time = 37.3 s Response Time = 23.38 sPhase 1 Phase 2

Without soCloud elasticity, the Response Time = 65.90 sNo request has failed

�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

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

Summary

• Reactivity face: ✤ Failures (High-availability) ✤ Flash crowd effect (Elasticity)

• Negligible Overhead introduced.

Outline

�83


2.Challenges

3.State of the art

4.Contributions

4.1.soCloud Model


5.Validation


�84

ConclusionsoCloud 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

PerspectivesShort-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

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)

socloud: distributed multi-cloud platform for deploying ... · socloud: distributed multi-cloud...

Documents