paraiso cloud2016.key
TRANSCRIPT
University of Lille & Inria Lille - Nord Europe (France)
Fawaz Paraïso | Stéphanie Challita | Yahya Al-Dhuraibi | Philippe Merle 9th International Conference on Cloud Computing(CLOUD 2016)
Model-Driven Management of Docker Containers
Jun27-July2,2016.SanFrancisco
Containers Technology
3
ContainersaddnewefficiencytoCloudComputing
Dockeristhewellknowcontainerstechnology
DockerAdoptionIsNotWithoutConcern…
Portability
lowresourceconsumption
Lightweight
Click to edit Master subtitle style
Model-Driven Management of Docker Containers
2 | Challenges for Managing Docker Containers
Jun27-July2,2016.SanFrancisco
A distributed Event Processing Application
5
Word
Word
Count
Count
Count
Count
Rolling
Rolling
Ranking
Rolling
RankingRolling
Jun27-July2,2016.SanFrancisco
Deployability of Application
6
Deployment
CorrectionsThisoperationcanberepeatedseveralstime
ResourceconsumedTimeconsumed
Application
Runtime
Appli
Currently,withDockerthereisnowaytoguaranteethedeployabilityofapplication
Jun27-July2,2016.SanFrancisco
Challenges for Managing Docker Containers
1. Challenge 1: Lack of the verification
7
Jun27-July2,2016.SanFrancisco
Challenges for Managing Docker Containers
1. Lack of the verification
2. Challenge 2: Resource management at runtime
9
Jun27-July2,2016.SanFrancisco
Synchronization between Design Time and Runtime
10
Runtime
AppDesignTime
Jun27-July2,2016.SanFrancisco
Challenges for Managing Docker Containers
1. Lack of the verification
2. Resources management at runtime
3. Challenge 3: Synchronization between design time and runtime
11
Jun27-July2,2016.SanFrancisco
Inconsistency Use of Containers Across Organizations
12
Theproblemof:MaintainabilityremainsInconsistency
Swarm Compose
Jun27-July2,2016.SanFrancisco
Challenges for Managing Docker Containers
1. Lack of the verification
2. Resources management at runtime
3. Synchronization between design time and runtime
4. Challenge 4: Inconsistency use of containers across
organizations
13
Jun27-July2,2016.SanFrancisco
Outline
14
1.Context 2.Problem Statement 3.Our approach 4.Evaluation 5.Conclusion 6. Future Work
Jun27-July2,2016.SanFrancisco
Model-Driven Engineering (MDE)
16
MDE=Abstraction+Automation
ModelisanabstractionthatenablestodealwithcomplexrealityinsimplifiedwayAutomationisamodeltransformationthatmutatesonmodeltoanother
ViewpointV2
M2
ViewpointV1
M1
ObjectConstraintLanguage(OCL)
Jun27-July2,2016.SanFrancisco
Our Approach
17
Model-DrivenManagementofDockerContainers
Oursolutionisfourphasemodel-driventool1. Analysisofconsistency2. Resourcemanagementatruntime2.SynchronizationofdesignedanddeployedContainers3.DesignTool
Jun27-July2,2016.SanFrancisco
Architecture Overview
18
Docker Model
Runtime
Legend:
Docker artifact
Model element
Connector
Generate artifacts
Update model elements
Connector
Jun27-July2,2016.SanFrancisco
Docker Model (1/2)‣ In MDE everything is a model
➡Container, Link, Host,Volume, etc… • All modeling artefacts in MDE are interrelated
19
Transformations
DockerFile,compose,swarm
Validation
Design
Jun27-July2,2016.SanFrancisco
Docker Container Architectural Constraints
21
Container-a Container-b
Thisarchitecturewillfailduringitsdeploymentduetothecyclebetweencontainers
context Container inv NoCycleBetweenContainerLinks:links->select(oclIsTypeOf(Link)).target->closure(links->select(oclIsTypeOf(Link)).target)->excludes(self)
OCLconstraint
Jun27-July2,2016.SanFrancisco
Synchronization
22
Docker Model
Runtime
Connector
Twowaysynchronization
Jun27-July2,2016.SanFrancisco
Connecting Docker Model Online
23
Docker Model
Runtime
Detectchanges Connector
DESIG
NTIM
ERU
NTIM
E
Model
Modelupdate
ChangeDetection
Monitoring
Jun27-July2,2016.SanFrancisco
Use Case
26
Storm topology
ranking bolt
rolling count rolling count
count count count
word spout 1 word spout 2
20VMswith8DockercontainersdeployedonaprivateCloud
Jun27-July2,2016.SanFrancisco
Overhead
27
Createaction Avg.starttime DockerModelOverheadDocker 168.509sec -DockerwithModel 170.382sec 1,11%
Startaction Avg.starttime DockerModelOverheadDocker 5.033sec -DockerwithModel 5.04sec 2,1%
Stopaction Avg.starttime DockerModelOverheadDocker 84.12sec -DockerwithModel 86.01sec 2,25%
Containercreationtime
Containerstartingtime
Containerstoppingtime
TheexperimentshowsthattheoverheadintroducedbyDockerModelisnegligibleregardingalladvantagesprovidedbyourapproach
8containerswereusedtoperformthisevaluation
Jun27-July2,2016.SanFrancisco
Scalability of Docker Model at Runtime
28
Totalof50,000containerswheregenerated.Foreachgeneration,themodelperformscreationandupdateactions.
Theaveragetimetakentogenerate50containersis14.30seconds
Jun27-July2,2016.SanFrancisco
ConclusionPresented an ongoing Model-Driven Management of
Docker Containers MDE tooling enables to design, reason, and deploy
containers Our approach provides to the users the possibility to
verify the deployability of Docker containers at design time Tested our approach on a representative application
scenario and deployed it to cloud
30
Jun27-July2,2016.SanFrancisco
Future WorkInvestigate the adaptation of managed containers as a
set of atomic changes
Extended our approach to other container solutions
32
Questions & comments
University of Lille & Inria Lille - Nord Europe (France)
Thank you!
http://occiware.org
FawazParaïso:[email protected]