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
Cloud computing in nutshell Virtualization On-demand Pay-per-use Elasticity 2
Context and motivation Application fil rouge Compute% View% Storage% Developer Go to the Cloud! 3
So many problems! Vendor Lock-in Geo-location Cloud-specific services Failures 4
solution: Multi-Cloud Do not put all your eggs in one basket Multi-Cloud Why not applying this precept of caution for cloud computing? 5
What is Multi-Cloud? Definition Multi-Cloud Computing using multiple cloud providers independent no agreement between providers 6
Multi-Cloud Context and motivation 74% Enterprises have a Multi-Cloud strategies 7
Towards Context Multi-Cloud and motivation Computing Multi-cloud is supposed to be the solution but 8
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 9
Challenges Multi-cloud Portability Multi-cloud Provisioning Multi-cloud Elasticity Multi-cloud High-Availability 10
Challenges Multi-cloud Portability Multi-cloud Provisioning Multi-cloud Elasticity Multi-cloud High availability Write once, deploy anywhere " without any modification App App Your application/data App App App 11
Challenges Multi-cloud Portability Multi-cloud Provisioning Multi-cloud Elasticity Multi-cloud High availability Resources App Deployment 12
Challenges Multi-cloud Portability Multi-cloud Provisioning Multi-cloud Elasticity Multi-cloud High availability Provisioning Resources Provisioning Provisioning 13
Multi-cloud Portability Challenges Multi-cloud Provisioning App deployment Multi-cloud Elasticity Multi-cloud High availability 14
Challenges Multi-cloud Portability Multi-cloud Provisioning Multi-cloud Elasticity Multi-cloud High availability 15
Challenges Multi-cloud Portability Multi-cloud Provisioning Multi-cloud Elasticity Multi-cloud High availability HA 16
Challenges Multi-cloud Portability Multi-cloud Provisioning Multi-cloud Elasticity Multi-cloud High availability HA HA HA 17
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 18
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 + + + + 19
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 + + + + 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
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
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 23
socloud Overview EC2 Developer distributed applications socloud Model socloud" Platform Azure Heroku 24
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 25
Objectives Provides a model to design a distributed applications in a simple and concise manner for a Multi-Cloud environment 26
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 27
socloud Model Extended SCA Model 28
SCA Service Component Architecture (SCA) Set of OASIS specifications Distributed applications Using SOA Compute View Storage Legend Composite Service Property Component Reference Wire SCA Contribution ZIP File 29
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
socloud Model based on SCA Why extend SCA model? Annota&on' +name& Implementa)on.Contribu)on1 +name& +value& Extension 1 Extension 2 31
socloud Model: implementation Implementa)on.Contribu)on1 +name& Extension 1 Provides high level conceptual view to a component Allows the deployment of the component as execution unit component Structured components of distributed applications 32
socloud Model: annotations Annota&on' @ +name& component +value& Allowing to associate non-functional requirements to a component The SCA model does not allow us to take into account these non-functional requirements Extension 2 33
socloud Model: annotations Annotation +name +value Placement " annotation Execution " annotation Availability " annotation Elasticity " annotation location closer vm database replication elasticity service 34
socloud Model: annotations Annotation +name +value Placement " annotation Execution " annotation Availability" annotation Elasticity " annotation location closer vm database replication elasticity service 35
socloud Model: annotations Placement " annotation location closer 1@location = value " 2 @closer = value 36
socloud Model: annotations 1@location = value Clouds provider Any Amazon EC2 Windows Azure Africa America Asia Europe Oceania (1) South Africa New york California Singapour Amazon Irland Irland France Australia (2) Paris Roubaix (3) 37
socloud Model: annotations 2 @closer = value @closer = C2 C1 C2 Latency 38
socloud Model: annotations Annotation +name +value Placement " annotation Execution " annotation Availability " annotation Elasticity " annotation location closer vm database replication elasticity service 39
socloud Model: annotations Execution " annotation 1 @vm = type_vm " 2 @database = name -> version " 3 @service = name -> version vm database Example Optional service 1 @vm = micro " 2 @database = MySQL " 3 @service = IronMQ -> 2.8.9 40
socloud Model: annotations Annotation +name +value Placement " annotation Execution " annotation Availability " annotation Elasticity " annotation location closer vm database replication elasticity service 41
socloud Model: annotations 1 @replication = number Availability " annotation Example replication @replication=5 C become LB C 42
socloud Model: annotations Annotation +name +value Placement " annotation Execution " annotation Availability " annotation Elasticity " annotation location closer vm database replication elasticity service 43
socloud Model: annotations 1 @elasticity = description Elasticity " annotation C scale in scale out elasticity C A DSL for describing elasticity 44
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 45
socloud Model: elasticity language Trigger scaling in 5 at ( 20:00 Friday) 46
socloud Model: annotations 3-Tiers Application @closer= Storage " @vm= xlarge->ubuntu @elasticity= Scaling in " When ( totalcost(computecost, 24 h) > 200 900 ) Compute " @database= MySQL @location= France " @location= France " View Storage 47
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) 48
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 49
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 50
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 self management > Autonomic is flexible > Component-based Reactive, flexible and self management platform 51
socloud Platform socloud Platform is a distributed component-based PaaS for managing Portability Provisioning Elasticity High-availability 52 Components
socloud platform high level view User socloud " master socloud agent socloud applications 53
socloud Platform: Multi-Cloud centric Architecture Trend in the socloud Platform Architecture single Cloud centric Architecture Multi-Cloud centric Architecture socloud master socloud master socloud master socloud agent socloud agent socloud agent socloud agent 54
socloud Platform detail view Load Balancer socloud master Service " Deployer Constraint " Validator Controller Node Provisioning PaaS Deployment Workload " Manager SaaS Deployment Monitoring socloud agent 55
socloud Platform: Fault Tolerance Let it Crash Application level Platform level Replication in different clouds Replication in different clouds 56
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 57
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 58
socloud Platform: deployment stack master agent socloud IaaS SCA container Servlet container Java runtime Linux/OS Resources 59
socloud Platform: deployment stack master SCA container agent socloud PaaS Servlet container 60
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. 61
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 62
Validation socloud Model socloud Platform 63
Validation: socloud Model Modeling of three concrete applications using the socloud Model 1. APISENSE application 2. DiCEPE application 3. P2P Monitoring application 64
Validation: socloud Model 1. APISENSE application [Nicolas Haderer] Create task Download sensing task Reward Publish task Subscribe task Sensing'Node' APISENSE' Central(Node( Subscribe task Publish task Sensing'Node' store data Sensing'Storage' Send data Send data Send data Send data 65
Validation: socloud Model Geo-location Paris Unpredictable growth of smartphones Availability despite failures Cost control 66
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> 67
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. 68
Validation: socloud Platform 1. Portability 2. High-availability 3. Elasticity 4. Overhead introduced by socloud 69
Validation: socloud Platform Portability 70
Validation: socloud Platform Deployed on 10 Clouds IaaS and PaaS 71
Validation: socloud Platform High-availability 72
Validation: socloud Platform MTBF* Availability= MTBF + MTTR** [Marcus et. al.] MTBF* = Mean Time Between Failure MTTR** = Mean Time To Recover [Marcus et. al.] : Blueprints for High availability 73
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 74
Validation: socloud Platform If it is assumed that a failure occurs once per year MTBF = 8760 Hours socloud Availability 8760 8760 + 0.06 = 99.999% Public clouds 8760 8760 + 7.5 = 99.914% 75
Validation: socloud Platform Elasticity 76
Flash crowd effect 3-Tiers application was deployed on ten cloud providers 60000" Phase 1 Phase 2 50000" Number%of%requests% 40000" 30000" 20000" 10000" 0" 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)% (a)" Total Number of Request = 3020000 77
Flash crowd effect without socloud elasticity 4500" Number'of'failed'requests' 4000" 3500" 3000" 2500" 2000" 1500" 1000" 500" 0" 0" 20" 40" 60" 80" 100" 120" 140" 160" 180" 200" Time'(slot"of"10"seconds)' 1.3% of requests are failed that correspond to 34039 Response Time = 65.90 s 78
Flash crowd effect with socloud elasticity 70" 60" Response'(me'(s)' 50" 40" 30" 20" Applica(on'replica(on' Balance'load' 10" 0" 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)' No request has failed Without socloud elasticity, the Response Time = 65.90 s Response Time = 37.3 s Response Time = 23.38 s Phase 1 Phase 2 79
Overhead introduced by socloud APP APP Implementation Execution time Overhead introduced by socloud (Application + FraSCAti) 10.85 sec - (Application + FraSCAti + socloud) 11,10 sec 2.3% 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% The benefit provided by the socloud Platform outweighs the difference in the execution time 81
Summary Reactivity face: Failures (High-availability) Flash crowd effect (Elasticity) Negligible Overhead introduced. 82
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 83
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. socloud Platform 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. 84
Perspectives Short-term further work The high-availability management despite software bugs. The elasticity management using reinforcement learning. Further Research Directions Security for Multi-Cloud. Sharing state between replicates. Take into account changes of the underlying platforms. 85
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) 86