Search-Based Genetic Optimization for Deployment and Reconfiguration - - PowerPoint PPT Presentation

▶
search based genetic optimization for deployment and
SMART_READER_LITE
LIVE PREVIEW

Search-Based Genetic Optimization for Deployment and Reconfiguration - - PowerPoint PPT Presentation

Sep 13, 2022 259 likes •676 views

Search-Based Genetic Optimization for Deployment and Reconfiguration of Software in the Cloud 35th International Conference on Software Engineering (ICSE 2013) San Francisco, CA, USA Sren Frey, Florian Fittkau, and Wilhelm Hasselbring

slide-1
SLIDE 1

Search-Based Genetic Optimization for Deployment and Reconfiguration of Software in the Cloud

35th International Conference on Software Engineering (ICSE 2013) San Francisco, CA, USA Sören Frey, Florian Fittkau, and Wilhelm Hasselbring

Software Engineering Group

  • Dept. Computer Science

Kiel University, Kiel, Germany May 23, 2013

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 1 / 20

slide-2
SLIDE 2

Migrating Software to the Cloud

Huge Variety of Cloud Deployment Options

Motivation

“Deploying and migrating software applications to the cloud involves many combinations of

  • ptions that vary widely in their

characteristics and performances, from different combinations of CPU, memory, storage, and network options to IT resource management services, to algorithms that can perform dynamic resource scaling. Developing software systems and choosing the most suitable cloud deployment option using these heterogeneous resources is both nontrivial and difficult [...]” ⇒ Impractical to incorporate all cloud environments in an actual comparison. IEEE Software March/ April 2012

[Grundy et al., 2012]

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 2 / 20

slide-3
SLIDE 3

Simple Cloud Deployment Option Example

(No reconfiguration rules, for simplicity)

Motivation

On premise deployment (Status quo deployment)

On-Premise Server 0 <<component>> Functionality A <<component>> Functionality C <<component>> Functionality B On-Premise Server 1 <<component>> Functionality E <<component>> Functionality D On-Premise Server 2 <<component>> Functionality F <<component>> Functionality G <<component>> Functionality H

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 3 / 20

slide-4
SLIDE 4

Simple Cloud Deployment Option Example

(No reconfiguration rules, for simplicity)

Motivation

On premise deployment Cloud Deployment Options (CDOs): (Status quo deployment) Cloud environment to use

On-Premise Server 0 <<component>> Functionality A <<component>> Functionality C <<component>> Functionality B On-Premise Server 1 <<component>> Functionality E <<component>> Functionality D On-Premise Server 2 <<component>> Functionality F <<component>> Functionality G <<component>> Functionality H

Cloud Environment?

  • Amazon EC2
  • BitRefinery
  • CloudSigma
  • Enomaly
  • GoGrid
  • Hosting.com
  • HP Cloud Compute
  • IBM SmartCloud
  • Joyent
  • Lunacloud
  • NephoScale
  • OpSource
  • Rackspace
  • ReliaCloud
  • Softlayer
  • Terremark Enterprise Cloud

...

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 3 / 20

slide-5
SLIDE 5

Simple Cloud Deployment Option Example

(No reconfiguration rules, for simplicity)

Motivation

On premise deployment Cloud Deployment Options (CDOs): (Status quo deployment) By chance

On-Premise Server 0 <<component>> Functionality A <<component>> Functionality C <<component>> Functionality B On-Premise Server 1 <<component>> Functionality E <<component>> Functionality D On-Premise Server 2 <<component>> Functionality F <<component>> Functionality G <<component>> Functionality H

Cloud Environment?

  • Amazon EC2
  • BitRefinery
  • CloudSigma
  • Enomaly
  • GoGrid
  • Hosting.com
  • HP Cloud Compute
  • IBM SmartCloud
  • Joyent
  • Lunacloud
  • NephoScale
  • OpSource
  • Rackspace
  • ReliaCloud
  • Softlayer
  • Terremark Enterprise Cloud

...

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 3 / 20

slide-6
SLIDE 6

Simple Cloud Deployment Option Example

(No reconfiguration rules, for simplicity)

Motivation

On premise deployment Cloud Deployment Options (CDOs): (Status quo deployment) Number of VM instances

On-Premise Server 0 <<component>> Functionality A <<component>> Functionality C <<component>> Functionality B On-Premise Server 1 <<component>> Functionality E <<component>> Functionality D On-Premise Server 2 <<component>> Functionality F <<component>> Functionality G <<component>> Functionality H <<VM instance>> <<VM instance>>

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 3 / 20

slide-7
SLIDE 7

Simple Cloud Deployment Option Example

(No reconfiguration rules, for simplicity)

Motivation

On premise deployment Cloud Deployment Options (CDOs): (Status quo deployment) VM instance types to use

On-Premise Server 0 <<component>> Functionality A <<component>> Functionality C <<component>> Functionality B On-Premise Server 1 <<component>> Functionality E <<component>> Functionality D On-Premise Server 2 <<component>> Functionality F <<component>> Functionality G <<component>> Functionality H <<VM instance>> m1.large <<VM instance>> m1.medium

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 3 / 20

slide-8
SLIDE 8

Simple Cloud Deployment Option Example

(No reconfiguration rules, for simplicity)

Motivation

On premise deployment Cloud Deployment Options (CDOs): (Status quo deployment) Mapping of components to VM images

On-Premise Server 0 <<component>> Functionality A <<component>> Functionality C <<component>> Functionality B On-Premise Server 1 <<component>> Functionality E <<component>> Functionality D On-Premise Server 2 <<component>> Functionality F <<component>> Functionality G <<component>> Functionality H <<VM instance>> m1.large

<<component>> Service 1 <<component>> Service 2

<<VM instance>> m1.medium

<<component>> Service 0

<<component>> Functionality E <<component>> Functionality D <<component>> Functionality F <<component>> Functionality G <<component>> Functionality H <<component>> Functionality A <<component>> Functionality C <<component>> Functionality B

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 3 / 20

slide-9
SLIDE 9

Huge Variety of Cloud Deployment Options

Motivation

So, which Cloud Deployment Option (CDO) to use? A CDO defines:

  • Cloud environment to use
  • Number of VM instances to start
  • VM instance types to use
  • Mapping of software components to VM images
  • Runtime reconfiguration rules
  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 4 / 20

slide-10
SLIDE 10

Huge Variety of Cloud Deployment Options

Motivation

So, which Cloud Deployment Option (CDO) to use? A CDO defines:

  • Cloud environment to use
  • Number of VM instances to start
  • VM instance types to use
  • Mapping of software components to VM images
  • Runtime reconfiguration rules

Example runtime reconfiguration rule: Start R1 = 2 new VM instances of VM instance type R2 = m1.medium if average CPU utilization ≥ R3 = 80% for R4 = 5 minutes.

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 4 / 20

slide-11
SLIDE 11

Huge Variety of Cloud Deployment Options

Motivation

So, which Cloud Deployment Option (CDO) to use? A CDO defines:

  • Cloud environment to use
  • Number of VM instances to start
  • VM instance types to use
  • Mapping of software components to VM images
  • Runtime reconfiguration rules

Example runtime reconfiguration rule: Start R1 = 2 new VM instances of VM instance type R2 = m1.medium if average CPU utilization ≥ R3 = 80% for R4 = 5 minutes. Which are the best suited values for R1, R2, R3, and R4?

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 4 / 20

slide-12
SLIDE 12

Outline

Motivation

1

Motivation

2

Project Context

3

CDOXplorer

4

Experiments

5

Conclusion

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 5 / 20

slide-13
SLIDE 13

Outline

Project Context

1

Motivation

2

Project Context

3

CDOXplorer

4

Experiments

5

Conclusion

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 5 / 20

slide-14
SLIDE 14

The CloudMIG Method

[Frey and Hasselbring, 2011]

Project Context

Existing System

A2

Actual Architecture

A1

Utilization Model Cloud Environment Model Target Architecture Mapping Model

A1 ? ?

Constraint Violations

A3 A4,A3 A5

Rating

A B C A6

Migrated System

A4,A3

Legend: A1: Extraction A2: Selection A3: Generation A4: Adaptation A5: Evaluation A6: Transformation Optional Mandatory

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 5 / 20

slide-15
SLIDE 15

CloudMIG and CloudMIG Xpress

❤tt♣✿✴✴✇✇✇✳❝❧♦✉❞♠✐❣✳♦r❣

Project Context

  • CDOXplorer and CDOSim: Part of our cloud migration

approach CloudMIG

  • Tool support for CloudMIG: Integration of CDOSim in

CloudMIG Xpress (based on [Fittkau et al., 2012a]) CloudMIG Xpress

CDOSim/ Benchmark components Data flow Basic data user needs to provide for simulating cloud deployment options Data only needed for dynamic and hybrid instruction counting approach

CloudMIG Xpress KDM Model Workload Profile Cloud Profile Mapping Model

(from sources)

Status Quo Deployment Node Software System

Monitoring Log Data

MIPIPS and Weights Benchmark <<Cloud provider X>> <<VM instance type Y>> VM Instance MIPIPS and Weights Benchmark CDOSim Instruction Counting Approaches Hybrid Static Dynamic

enrich

CDOXplorer

fitness values

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 6 / 20

slide-16
SLIDE 16

Simulating Cloud Deployment Options (CDOs)

CDOSim [Fittkau et al., 2012a,b]

Project Context

  • Our simulator CDOSim can simulate CDOs and identify costs, response times, and #SLA

violations

  • Example from previous case study:
  • JPetStore deployed to Amazon EC2 and private Eucalyptus cloud
  • Start new VM instances when CPU utilization > 70% for at least one minute
  • Shut down VM instance when CPU utilization < 30% for at least one minute
  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 7 / 20

slide-17
SLIDE 17

Simulating Cloud Deployment Options (CDOs)

CDOSim [Fittkau et al., 2012a,b]

Project Context

  • Our simulator CDOSim can simulate CDOs and identify costs, response times, and #SLA

violations

  • Example from previous case study:
  • JPetStore deployed to Amazon EC2 and private Eucalyptus cloud
  • Start new VM instances when CPU utilization > 70% for at least one minute
  • Shut down VM instance when CPU utilization < 30% for at least one minute

Average CPU Utilization

Experiment time [day hour:minute] 01 00:00 01 07:00 01 14:00 01 21:00 10 30 50 70 90 Average CPU utilization over all allocated nodes [%] 1 2 3 4 5 6 7 8 Number of allocated nodes Average CPU utilization Number of allocated nodes

Figure : Simulated CPU utilization

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 7 / 20

slide-18
SLIDE 18

Simulating Cloud Deployment Options (CDOs)

CDOSim [Fittkau et al., 2012a,b]

Project Context

  • Our simulator CDOSim can simulate CDOs and identify costs, response times, and #SLA

violations

  • Example from previous case study:
  • JPetStore deployed to Amazon EC2 and private Eucalyptus cloud
  • Start new VM instances when CPU utilization > 70% for at least one minute
  • Shut down VM instance when CPU utilization < 30% for at least one minute

Average CPU Utilization

Experiment time [day hour:minute] 01 00:00 01 07:00 01 14:00 01 21:00 10 30 50 70 90 Average CPU utilization over all allocated nodes [%] 1 2 3 4 5 6 7 8 Number of allocated nodes Average CPU utilization Number of allocated nodes

Figure : Simulated CPU utilization

Average CPU Utilization

Experiment time [day hour:minute] 01 00:00 01 07:00 01 14:00 01 21:00 10 30 50 70 90 Average CPU utilization over all allocated nodes [%] 1 2 3 4 5 6 7 8 Number of allocated nodes Average CPU utilization Number of allocated nodes

Figure : Measured CPU utilization

⇒ High precision ⇒ But: Huge number of possible CDOs ⇒ Simulating all CDOs (sequentially) would last for years

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 7 / 20

slide-19
SLIDE 19

Outline

CDOXplorer

1

Motivation

2

Project Context

3

CDOXplorer

4

Experiments

5

Conclusion

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 8 / 20

slide-20
SLIDE 20

CDO Optimization Approach

CDOXplorer

Goal:

  • Support comparison of cloud deployment options
  • Find suitable trade-offs between cost, response times,

and SLA violations (number of method timeouts)

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 8 / 20

slide-21
SLIDE 21

CDO Optimization Approach

CDOXplorer

Goal:

  • Support comparison of cloud deployment options
  • Find suitable trade-offs between cost, response times,

and SLA violations (number of method timeouts) Challenges:

  • Optimization problem with multiple objectives
  • SLA-aware service deployment optimization is

NP-hard [Canfora et al., 2005]

  • Complex and non-linear correlations of input

parameters

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 8 / 20

slide-22
SLIDE 22

CDO Optimization Approach

CDOXplorer

Goal:

  • Support comparison of cloud deployment options
  • Find suitable trade-offs between cost, response times,

and SLA violations (number of method timeouts) Challenges:

  • Optimization problem with multiple objectives
  • SLA-aware service deployment optimization is

NP-hard [Canfora et al., 2005]

  • Complex and non-linear correlations of input

parameters Approach:

  • Explore design space with a genetic algorithm called

CDOXplorer

  • Use our simulator CDOSim to compute fitness of CDOs

(⇒ simulation-based optimization)

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 8 / 20

slide-23
SLIDE 23

CDOXplorer Overview

CDOXplorer

  • IaaS cloud environments (basic building blocks: virtual machines)
  • Usage of actual workload data (e.g., from monitoring log files)
  • Optimizes runtime reconfiguration rules at design time

(migration planning)

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 9 / 20

slide-24
SLIDE 24

CDOXplorer Overview

CDOXplorer

  • IaaS cloud environments (basic building blocks: virtual machines)
  • Usage of actual workload data (e.g., from monitoring log files)
  • Optimizes runtime reconfiguration rules at design time

(migration planning) Input and Output:

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 9 / 20

slide-25
SLIDE 25

Scaling Types

CDOXplorer

scale out/ scale in scale up/ scale down

<<VM instance>> m1.small (1) <<component>> Service 1 <<component>> Service n <<VM instance>> m1.small (2) <<component>> Service 1 <<component>> Service n

scale in scale out

<<VM instance>> m1.small (1) <<component>> Service 1 <<component>> Service n <<VM instance>> m1.small <<component>> Service 1 <<component>> Service n <<VM instance>> m2.4xlarge <<component>> Service 1 <<component>> Service n

scale down scale up

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 10 / 20

slide-26
SLIDE 26

Cloud Deployment Options

CDOXplorer

Structure of a Cloud Deployment Option (Phenotype)

Cloud Deployment Option Node Configuration

  • id

Cloud Environment Service Composition

  • vmInstanceTypeID
  • nrVMsToStart

Initial Start Config

  • minNrVMs
  • mipipsMultiple

Grow Rule

  • minNrVMs
  • mipipsMultiple

Shrink Rule

  • id

Service

  • scaleUp
  • scaleOut

<<enumeration>> Grow Action

  • scaleDown
  • scaleIn

<<enumeration>> Shrink Action

  • cpuUtilizationThreshold
  • timePeriod

Condition

  • singleVM
  • allVMs

<<enumeration>> Scope kdm:code:CodeModel kdm:code:Package * 1..* * 0..1 0..1 1..* * 1..* 1..*

. . .

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 11 / 20

slide-27
SLIDE 27

Compound Chromosome Overview

CDOXplorer

Structure of a Cloud Deployment Option (Genotype)

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 12 / 20

slide-28
SLIDE 28

Gene Design

CDOXplorer

  • Gene

Range Description Chromosome CE N Cloud environment id Cloud Environment SE N Service id Service Composition IT N VM Instance type id Initial Start Configuration NI N

  • Nr. of VM instances to start initially

Initial Start Configuration GA 0,1 Grow action; 0: scale up, 1: scale out Grow Rule G1 N Minimum nr. of VM instances Grow Rule G2 1.1-3.0 MIPIPS multiple in steps of 0.1 Grow Rule G3 0,1 Condition scope; 0: single VM, 1: all VMs Grow Rule G4 0.05-1.0 Condition median utilization in steps of 0.05 Grow Rule G5 5-60 Condition time period in steps of 5 minutes Grow Rule SA 0,1 Shrink action; 0: scale down, 1: scale in Shrink Rule S1 N Minimum nr. of VM instances Shrink Rule S2 0.1-0.9 MIPIPS multiple in steps of 0.1 Shrink Rule S3 0,1 Condition scope; 0: single VM, 1: all VMs Shrink Rule S4 0.0-0.95 Condition median utilization in steps of 0.05 Shrink Rule S5 5-60 Condition time period in steps of 5 minutes Shrink Rule

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 13 / 20

slide-29
SLIDE 29

Cloud Deployment Options - Examples

CDOXplorer

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 14 / 20

slide-30
SLIDE 30

Cloud Deployment Options - Examples

CDOXplorer

  • ⇒
  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 14 / 20

slide-31
SLIDE 31

Crossover Operator

CDOXplorer

4 sub crossover operators corresponding to 4 crossover points

Crossover point Sub operator Description CP1 CE Swap cloud environments CP2 CI Swap initial start configurations CP3 CG Swap grow rules CP4 CS Swap shrink rules

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 15 / 20

slide-32
SLIDE 32

Crossover Operator

CDOXplorer

4 sub crossover operators corresponding to 4 crossover points

Crossover point Sub operator Description CP1 CE Swap cloud environments CP2 CI Swap initial start configurations CP3 CG Swap grow rules CP4 CS Swap shrink rules

Example: CI

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 15 / 20

slide-33
SLIDE 33

Outline

Experiments

1

Motivation

2

Project Context

3

CDOXplorer

4

Experiments

5

Conclusion

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 16 / 20

slide-34
SLIDE 34

Case Study: Apache OfBiz

Experiments

Experimental Setting:

  • Fitness function: Simulate CDOs of

Apache OfBiz 10.04 (❤tt♣✿✴✴♦❢❜✐③✳❛♣❛❝❤❡✳♦r❣✴)

  • Single-cloud scenario (SCS): Amazon EC2
  • Multi-cloud scenario (SCM): Amazon EC2,

Microsoft Windows Azure, Eucalyptus

  • Comparing CDOXplorer with three

alternative search methods:

  • SI-RS: Simple random sampling
  • SY-RS: Systematic random sampling
  • SI-AN: Simulated annealing

Metrics Hypervolume (❍❱) and Inverted Generational Distance (■●❉):

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 16 / 20

slide-35
SLIDE 35

Case Study: Apache OfBiz (cont’d)

Experiments

Baseline (best-known) pareto-optimal front for SCS:

20 40 60 80 100 120 50000 100000 150000 200000 250000 300000 500 1000 1500 2000 2500 3000 3500

Cost [$] Response times [ms] SLA violations

  • ●
  • ●
  • ● ●
  • ●
  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 17 / 20

slide-36
SLIDE 36

Case Study: Apache OfBiz (cont’d)

Experiments

Baseline (best-known) pareto-optimal front for SCS:

20 40 60 80 100 120 50000 100000 150000 200000 250000 300000 500 1000 1500 2000 2500 3000 3500

Cost [$] Response times [ms] SLA violations

  • ●
  • ●
  • ● ●
  • ●
  • 500

1000 1500 2000 2500 3000 3500 50000 100000 150000 200000 250000 300000 20 40 60 80 100 120

Response times [ms] Cost [$] SLA violations

  • ●
  • ●
  • ●
  • ●
  • ●
  • ●
  • ●
  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 17 / 20

slide-37
SLIDE 37

Case Study: Apache OfBiz (cont’d)

Experiments

Exemplary results for SCS:

Search Method Metric CDOXplorer SI-RS SY-RS SI-AN I.G. Distance Mean 2.70E-02 3.67E-02 4.11E-02 3.28E-02 SD 2.10E-03 2.13E-03 3.61E-03 2.85E-03 Median 2.72E-02 3.65E-02 4.21E-02 3.20E-02 Min (best) 2.16E-02 3.34E-02 3.40E-02 2.76E-02 Max (worst) 3.03E-02 4.07E-02 4.83E-02 3.95E-02 Hypervolume Mean 4.48E-01 4.41E-01 4.41E-01 4.44E-01 SD 2.08E-03 1.96E-03 2.89E-03 2.09E-03 Median 4.48E-01 4.40E-01 4.41E-01 4.44E-01 Min (worst) 4.44E-01 4.36E-01 4.35E-01 4.40E-01 Max (best) 4.54E-01 4.46E-01 4.46E-01 4.48E-01

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 18 / 20

slide-38
SLIDE 38

Case Study: Apache OfBiz (cont’d)

Experiments

CDOXplorer advantage relative to other approaches:

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 19 / 20

slide-39
SLIDE 39

Outline

Conclusion

1

Motivation

2

Project Context

3

CDOXplorer

4

Experiments

5

Conclusion

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 20 / 20

slide-40
SLIDE 40

Conclusion

Conclusion

  • Context:
  • CDOs can be simulated with our simulator

CDOSim (estimation of future costs, response times, and SLA violations)

  • There exists a vast amount of CDOs,

simulating all (sequentially) lasts thousands of years

  • Approach:
  • Genetic algorithm CDOXplorer allows to find

best-suited CDOs (pareto optimal set)

  • Part of cloud migration approach CloudMIG
  • CDOSim constitutes fitness function

(simulation-based optimization)

  • Tool support:

CloudMIG Xpress ❤tt♣✿✴✴✇✇✇✳❝❧♦✉❞♠✐❣✳♦r❣

  • S. Frey, F. Fittkau, and W. Hasselbring

Optimizing the Deployment of Software in the Cloud May 23, 2013 20 / 20

slide-41
SLIDE 41

References

  • G. Canfora, M. Di Penta, R. Esposito, and M. L. Villani. An approach for QoS-aware service composition based
  • n genetic algorithms. In Proceedings of the 2005 conference on Genetic and evolutionary computation,

GECCO ’05, pages 1069–1075, New York, NY, USA, 2005. ACM. ISBN 1-59593-010-8. doi: ✶✵✳✶✶✹✺✴✶✵✻✽✵✵✾✳✶✵✻✽✶✽✾. F . Fittkau, S. Frey, and W. Hasselbring. CDOSim: simulating cloud deployment options for software migration

  • support. In IEEE 6th International Workshop on the Maintenance and Evolution of Service-Oriented and

Cloud-Based Systems (MESOCA), 2012, pages 37–46, sept. 2012a. doi: ✶✵✳✶✶✵✾✴▼❊❙❖❈❆✳✷✵✶✷✳✻✸✾✷✺✾✾. F . Fittkau, S. Frey, and W. Hasselbring. Cloud User-Centric Enhancements of the Simulator CloudSim to Improve Cloud Deployment Option Analysis. In F . De Paoli, E. Pimentel, and G. Zavattaro, editors, Proceedings of the European Conference on Service-Oriented and Cloud Computing (ESOCC), volume 7592 of Lecture Notes in Computer Science, pages 200–207. Springer Berlin / Heidelberg, 2012b. ISBN 978-3-642-33426-9. doi: ✶✵✳✶✵✵✼✴✾✼✽✲✸✲✻✹✷✲✸✸✹✷✼✲✻❴✶✺.

  • S. Frey and W. Hasselbring. The CloudMIG approach: Model-based migration of software systems to

cloud-optimized applications. International Journal on Advances in Software, 4(3 and 4):342–353, Apr.

  • 2011. URL ❤tt♣✿✴✴✇✇✇✳t❤✐♥❦♠✐♥❞✳♦r❣✴✐♥❞❡①✳♣❤♣❄✈✐❡✇❂❛rt✐❝❧❡✫❛rt✐❝❧❡✐❞❂s♦❢t❴✈✹❴♥✸✹❴✷✵✶✶❴✽.

(ISSN: 1942-2628).

  • J. Grundy, G. Kaefer, J. Keong, and A. Liu. Guest Editors’ Introduction: Software Engineering for the Cloud.

IEEE Software, 29:26–29, 2012. ISSN 0740-7459. doi: ✶✵✳✶✶✵✾✴▼❙✳✷✵✶✷✳✸✶.

Sören Frey The CloudMIG Approach

  • Apr. 11, 2013

21