feifei chen swinburne university of technology melbourne, australia automating performance and...

28
Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS Provider: 00111D | TOID:

Upload: dorothy-roberts

Post on 19-Jan-2016

221 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Feifei Chen

Swinburne University of Technology

Melbourne, Australia

Automating Performance and Energy Consumption Analysis for Cloud Applications

CRICOS Provider: 00111D | TOID: 3059

Page 2: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN

• Background• Problem Analysis• Approach• Evaluation

2

Outlines

Page 3: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN 3

Cloud Computing

Page 4: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN 4

Benefits of Cloud

Page 5: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN 5

Dark Side of Cloud

High CO2 emissions contribution

Gartner Report 2007: IT industry contributes 2% of world's total CO2 emissions

New York Times 2012: Data centres use about 30 billion watts of electricity per hour worldwide, equivalent to the output of about 30 nuclear power plants

Page 6: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN 6

Dark Side of CloudHigh Operational Cost

U.S. EPA Report 2007: 1.5% of total U.S. power consumption used by data centers which has more than doubled since 2000 and costs $4.5 billion

Page 7: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN

Green Cloud Computing

7

Page 8: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN

Service Level Agreement (SLA)

8

Page 9: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN 9

A key objective of cloud service providers:

Develop solutions to cloud application deployment and management with minimum energy consumption while still guaranteeing performance and other Service Level Agreement (SLA) targets.

Energy Consumption

PerformanceService Providers

Understand both system performance and energy

consumption pattern

Page 10: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN

• Background• Problem Analysis• Approach• Evaluation

10

Outlines

Page 11: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN

Understand both system performance and energy consumption pattern: • running extensive experiments with heterogeneous

parameters/metrics and workloads;• collecting appropriate cloud and application

energy/performance measurements;• performing energy/performance trade-off analysis.

11

Page 12: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN 12

JPetStore Deployment

1000 User

s

Browse Product5000User

s

Workload

Page 13: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN

• Manually performing the tasks is tedious and time-consuming

• Cloud system performance is related to• Architecture• Workload

13

Challenges

An automated performance and energy consumption evaluation framework is imperative

The framework should be able to accommodate different cloud system architectures and adopt

different application workloads during load tests and the trade-off evaluation process

Page 14: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN

• Background• Problem Analysis• Approach• Evaluation

14

Outlines

Page 15: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Text line

Automated Performance and energy analysis tool - StressCloud

Deploy Load Test Services

Deploy Load Test Services

Load Test Scripts

Model Cloud System Workload

Model Cloud System Workload

Stored ResultsStored Results

Performance Engineer

Model Cloud Architecture

Model Cloud Architecture

Workload Deployment Scripts

Generate Load Test Scripts

Generate Load Test Scripts

Generate Workload Deployment ScriptsGenerate Workload Deployment Scripts

Workload Model

Cloud System Architecture

Model

Apply Load Tests to Cloud

Apply Load Tests to Cloud

Test Results

Collect Performance and

Energy Data

Collect Performance and

Energy Data

Visualise ResultsVisualise Results

12

4

3

5

6

7

StressCLoud

Cloud

8

Page 16: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Text line

Automated Performance and energy analysis tool - StressCloud

Cloud Architecture Model:

All available resources in the target cloud system and their detailed configurations.

Page 17: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Text line

Automated Performance and energy analysis tool - StressCloud

Cloud Application Workload Model: A set of Tasks modelling the target cloud application behaviour

Computation-Intensive CPU-Intensive Memory-Intensive

Data-Intensive Communication-Intensive

Task: A stochastic form chart specifying the detailed user requests and required responses from the cloud system

Task Type Service Type in StressCloudCPU-intensive Fibonacci sequence calculatingMemory-intensive File processingData-intensive Rational database operatingCommunication-intensive HTTP request/response

Page 18: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Text line

High-level Workload Model of JPetStore

Page 19: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Text lineStochastic Form-Chart ExampleOf Each Task

Page 20: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Text line

Architecture ModelCloud System Architecture

Model Example

Page 21: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Text line

Workload Deployment

Scripts Example

Page 22: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Text lineLoad Testing

Scripts Example

Page 23: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Text lineVisualized Results

Page 24: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Swinburne

SCIENCE | TECHNOLOGY | INNOVATION | BUSINESS | DESIGN

• Background• Problem Analysis• Approach• Evaluation

24

Outlines

Page 25: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Text line

Experiment SetupEnergy and performance profiling framework

VM configuration 

Virtual Machine

Number of Cores RAM Hard Disk

  Small 1 2GB 80GB  Medium 2 4GB 80GB  Large 3 6GB 80GB

XLarge 4 8GB 80GB

Cloud Server

VirtualMachine VirtualMachine

Apache Tomcat

…...Load Test Web Service

MS SQL Server

Apache Tomcat

Load Test Web Service

MS SQL Server

Monitor and Load Generator

StressCloud

PowerNodeGreenWave Gateway

Router

Page 26: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Text line

Experiment ResultsTest set 1: Keep the resource allocation strategy constant

while changing workload

System Configurations: 1 Large VM (3CPUs and 6GB RAM)

Workload: user workload 50~200

Energy Consumption and Throughput

Page 27: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Text line

Experiment ResultsTest set 2: Keep the workload constant while changing the

resource allocation strategy

System Configuration: 1Large: Deploy three types of tasks on the 1 VM.3Small(D): Deploy three types of tasks on different VMs.3Small(S): Deploy three types of tasks on the same VM with workloads evenly distributed across all VMs.

Workload: user request 100.

Energy Consumption and Throughput

Page 28: Feifei Chen Swinburne University of Technology Melbourne, Australia Automating Performance and Energy Consumption Analysis for Cloud Applications CRICOS

Feifei Chen

[email protected]

Thanks!

CRICOS Provider: 00111D | TOID: 3059