{ Considering I/O Processing inCloudSim for Performance and

Energy Evaluation }

Hamza Ouarnoughi

Plan

Context

Introduction & Problem Statement

Contribution

Evaluation

Conclusion & Future Work

23/6/2016 2

Context

23/6/2016 3

Context

General Context

This work is part of a project that aims to optimizeperformance and energy consumption of a hybrid storagesystem in the context of IaaS Cloud

Storage Management Model

The storage management is an autonomic loop that obeys theMAPE-K (Monitor, Analyze, Plan, Execute - Knowledge) model :

MonitoringVM I/O Tracing

AnalyzeVM I/O Workload

Profiling

PlanVM Storage Cost

Modeling & Optimization

ExecuteVM & Storage

Migration Scheduling

KnowledgeI/O Profiles &

Placement Plan History

Cloud Infrastructure (Compute, Network, Storage)

23/6/2016 4

Context

General Context

This work is part of a project that aims to optimizeperformance and energy consumption of a hybrid storagesystem in the context of IaaS Cloud

Storage Management Model

The storage management is an autonomic loop that obeys theMAPE-K (Monitor, Analyze, Plan, Execute - Knowledge) model :

MonitoringVM I/O Tracing

AnalyzeVM I/O Workload

Profiling

PlanVM Storage Cost

Modeling & Optimization

ExecuteVM & Storage

Migration Scheduling

KnowledgeI/O Profiles &

Placement Plan History

Cloud Infrastructure (Compute, Network, Storage)

23/6/2016 4

Introduction & ProblemStatement

23/6/2016 5

Introduction & Problem Statement

Cloud Infrastructure Power Consumption

Minimizing infrastructure power consumption : animportant concern for Cloud providers

VM placement optimization methods : popular approachesto minimize data centers power consumption

State of the art

VM placement optimization : usually based on CPU load

Storage activities : may greatly contribute to the overalldata center power consumption (up to 40% [1])

CloudSim : the most used simulator for VM placementapproaches implementation and evaluation

[1] Z. Li, K. M. Greenan, A. W. Leung, E. Zadok ”Power Consumption in Enterprise-Scale Backup Storage Systems”

23/6/2016 6

Introduction & Problem Statement

Cloud Infrastructure Power Consumption

Minimizing infrastructure power consumption : animportant concern for Cloud providers

VM placement optimization methods : popular approachesto minimize data centers power consumption

State of the art

VM placement optimization : usually based on CPU load

Storage activities : may greatly contribute to the overalldata center power consumption (up to 40% [1])

CloudSim : the most used simulator for VM placementapproaches implementation and evaluation

[1] Z. Li, K. M. Greenan, A. W. Leung, E. Zadok ”Power Consumption in Enterprise-Scale Backup Storage Systems”

23/6/2016 6

Introduction & Problem Statement

Cloud Infrastructure Power Consumption

Minimizing infrastructure power consumption : animportant concern for Cloud providers

VM placement optimization methods : popular approachesto minimize data centers power consumption

State of the art

VM placement optimization : usually based on CPU load

Storage activities : may greatly contribute to the overalldata center power consumption (up to 40% [1])

CloudSim : the most used simulator for VM placementapproaches implementation and evaluation

[1] Z. Li, K. M. Greenan, A. W. Leung, E. Zadok ”Power Consumption in Enterprise-Scale Backup Storage Systems”

23/6/2016 6

Introduction & Problem Statement

Cloud Infrastructure Power Consumption

Minimizing infrastructure power consumption : animportant concern for Cloud providers

VM placement optimization methods : popular approachesto minimize data centers power consumption

State of the art

VM placement optimization : usually based on CPU load

Storage activities : may greatly contribute to the overalldata center power consumption (up to 40% [1])

CloudSim : the most used simulator for VM placementapproaches implementation and evaluation

[1] Z. Li, K. M. Greenan, A. W. Leung, E. Zadok ”Power Consumption in Enterprise-Scale Backup Storage Systems”

23/6/2016 6

Introduction & Problem Statement

Cloud Infrastructure Power Consumption

Minimizing infrastructure power consumption : animportant concern for Cloud providers

VM placement optimization methods : popular approachesto minimize data centers power consumption

State of the art

VM placement optimization : usually based on CPU load

Storage activities : may greatly contribute to the overalldata center power consumption (up to 40% [1])

CloudSim : the most used simulator for VM placementapproaches implementation and evaluation

[1] Z. Li, K. M. Greenan, A. W. Leung, E. Zadok ”Power Consumption in Enterprise-Scale Backup Storage Systems”

23/6/2016 6

Introduction & Problem Statement

Cloud Infrastructure Power Consumption

Minimizing infrastructure power consumption : animportant concern for Cloud providers

VM placement optimization methods : popular approachesto minimize data centers power consumption

State of the art

VM placement optimization : usually based on CPU load

Storage activities : may greatly contribute to the overalldata center power consumption (up to 40% [1])

CloudSim : the most used simulator for VM placementapproaches implementation and evaluation

[1] Z. Li, K. M. Greenan, A. W. Leung, E. Zadok ”Power Consumption in Enterprise-Scale Backup Storage Systems”

23/6/2016 6

Introduction & Problem Statement

Cloud Infrastructure Power Consumption

Minimizing infrastructure power consumption : animportant concern for Cloud providers

VM placement optimization methods : popular approachesto minimize data centers power consumption

State of the art

VM placement optimization : usually based on CPU load

Storage activities : may greatly contribute to the overalldata center power consumption (up to 40% [1])

CloudSim : the most used simulator for VM placementapproaches implementation and evaluation

[1] Z. Li, K. M. Greenan, A. W. Leung, E. Zadok ”Power Consumption in Enterprise-Scale Backup Storage Systems”

23/6/2016 6

Problem Statement : CloudSim

Power Consumption & I/O Workload Execution

During I/O workload execution, power is mainly consumed by :

Storage system components

CPU and memory when processing I/Os (i.e. executing theI/O software stack)

CloudSim & Storage Capabilities

VM I/O workload execution : not considered (i.e. itsrelated time and power consumption)

Storage system : limited and consists of an HDD-based SAN(Storage Area Network)

23/6/2016 7

Problem Statement : CloudSim

Power Consumption & I/O Workload Execution

During I/O workload execution, power is mainly consumed by :

Storage system components

CPU and memory when processing I/Os (i.e. executing theI/O software stack)

CloudSim & Storage Capabilities

VM I/O workload execution : not considered (i.e. itsrelated time and power consumption)

Storage system : limited and consists of an HDD-based SAN(Storage Area Network)

23/6/2016 7

Problem Statement : CloudSim

Power Consumption & I/O Workload Execution

During I/O workload execution, power is mainly consumed by :

Storage system components

CPU and memory when processing I/Os (i.e. executing theI/O software stack)

CloudSim & Storage Capabilities

VM I/O workload execution : not considered (i.e. itsrelated time and power consumption)

Storage system : limited and consists of an HDD-based SAN(Storage Area Network)

23/6/2016 7

Problem Statement : CloudSim

Power Consumption & I/O Workload Execution

During I/O workload execution, power is mainly consumed by :

Storage system components

CPU and memory when processing I/Os (i.e. executing theI/O software stack)

CloudSim & Storage Capabilities

VM I/O workload execution : not considered (i.e. itsrelated time and power consumption)

Storage system : limited and consists of an HDD-based SAN(Storage Area Network)

23/6/2016 7

Problem Statement : CloudSim

Power Consumption & I/O Workload Execution

During I/O workload execution, power is mainly consumed by :

Storage system components

CPU and memory when processing I/Os (i.e. executing theI/O software stack)

CloudSim & Storage Capabilities

VM I/O workload execution : not considered (i.e. itsrelated time and power consumption)

Storage system : limited and consists of an HDD-based SAN(Storage Area Network)

23/6/2016 7

Problem Statement : CloudSim

Power Consumption & I/O Workload Execution

During I/O workload execution, power is mainly consumed by :

Storage system components

CPU and memory when processing I/Os (i.e. executing theI/O software stack)

CloudSim & Storage Capabilities

VM I/O workload execution : not considered (i.e. itsrelated time and power consumption)

Storage system : limited and consists of an HDD-based SAN(Storage Area Network)

23/6/2016 7

Problem Statement : CloudSim

Power Consumption & I/O Workload Execution

During I/O workload execution, power is mainly consumed by :

Storage system components

CPU and memory when processing I/Os (i.e. executing theI/O software stack)

CloudSim & Storage Capabilities

VM I/O workload execution : not considered (i.e. itsrelated time and power consumption)

Storage system : limited and consists of an HDD-based SAN(Storage Area Network)

23/6/2016 7

Contribution

23/6/2016 8

Contribution

Contributions Summary

We propose an extension for CloudSim storage capabilities inform of three contributions :

Considering VMs I/O workload execution and related timeand power consumption

Adding Flash-based SSD and local storage support, andpower related capabilities

Adding a model to compute the CPU time related to I/Oworkload processing

23/6/2016 9

Contribution

Contributions Summary

We propose an extension for CloudSim storage capabilities inform of three contributions :

Considering VMs I/O workload execution and related timeand power consumption

Adding Flash-based SSD and local storage support, andpower related capabilities

Adding a model to compute the CPU time related to I/Oworkload processing

23/6/2016 9

Contribution

Contributions Summary

We propose an extension for CloudSim storage capabilities inform of three contributions :

Considering VMs I/O workload execution and related timeand power consumption

Adding Flash-based SSD and local storage support, andpower related capabilities

Adding a model to compute the CPU time related to I/Oworkload processing

23/6/2016 9

Contribution

Contributions Summary

We propose an extension for CloudSim storage capabilities inform of three contributions :

Considering VMs I/O workload execution and related timeand power consumption

Adding Flash-based SSD and local storage support, andpower related capabilities

Adding a model to compute the CPU time related to I/Oworkload processing

23/6/2016 9

Contribution

Contributions Summary

We propose an extension for CloudSim storage capabilities inform of three contributions :

Considering VMs I/O workload execution and related timeand power consumption

Adding Flash-based SSD and local storage support, andpower related capabilities

Adding a model to compute the CPU time related to I/Oworkload processing

23/6/2016 9

Time Computation

Before the Extension

CloudSim I/O activity time : only affected by theCloudlet 1transfer during its submission or migration

CPU Activity

I/O Activity

Time

Cloudlet input file creation

CPU utilization (Cloudlet instruction execution)

Execution time

...

Figure – Time model before the extension

1. In CloudSim, a Cloudlet is a process representing a CPU workload andexecuted by a VM

23/6/2016 10

Time Computation

Before the Extension

CloudSim I/O activity time : only affected by theCloudlet 1transfer during its submission or migration

CPU Activity

I/O Activity

Time

Cloudlet input file creation

CPU utilization (Cloudlet instruction execution)

Execution time

...

Figure – Time model before the extension

1. In CloudSim, a Cloudlet is a process representing a CPU workload andexecuted by a VM

23/6/2016 10

Time Computation

Before the Extension

CloudSim I/O activity time : only affected by theCloudlet 1transfer during its submission or migration

CPU Activity

I/O Activity

Time

Cloudlet input file creation

CPU utilization (Cloudlet instruction execution)

Execution time

...

Figure – Time model before the extension

1. In CloudSim, a Cloudlet is a process representing a CPU workload andexecuted by a VM

23/6/2016 10

Time computation

After the Extension

Our extension adds two events affecting the simulation time :

I/O workload execution (i.e. storage device and CPU time)

VM image (virtual disk) creation (e.g. copy from arepository to the local storage)

CPU Activity

I/O Activity

Execution time

...

...

VM Image creationCloudlet input file

creation

CloudletCPU

utilization

I/O related CPU

utilization

I/O execution

Time

Figure – Time model after the extension

23/6/2016 11

Time computation

After the Extension

Our extension adds two events affecting the simulation time :

I/O workload execution (i.e. storage device and CPU time)

VM image (virtual disk) creation (e.g. copy from arepository to the local storage)

CPU Activity

I/O Activity

Execution time

...

...

VM Image creationCloudlet input file

creation

CloudletCPU

utilization

I/O related CPU

utilization

I/O execution

Time

Figure – Time model after the extension

23/6/2016 11

Time computation

After the Extension

Our extension adds two events affecting the simulation time :

I/O workload execution (i.e. storage device and CPU time)

VM image (virtual disk) creation (e.g. copy from arepository to the local storage)

CPU Activity

I/O Activity

Execution time

...

...

VM Image creationCloudlet input file

creation

CloudletCPU

utilization

I/O related CPU

utilization

I/O execution

Time

Figure – Time model after the extension

23/6/2016 11

Time computation

After the Extension

Our extension adds two events affecting the simulation time :

I/O workload execution (i.e. storage device and CPU time)

VM image (virtual disk) creation (e.g. copy from arepository to the local storage)

CPU Activity

I/O Activity

Execution time

...

...

VM Image creationCloudlet input file

creation

CloudletCPU

utilization

I/O related CPU

utilization

I/O execution

Time

Figure – Time model after the extension

23/6/2016 11

Time computation

After the Extension

Our extension adds two events affecting the simulation time :

I/O workload execution (i.e. storage device and CPU time)

VM image (virtual disk) creation (e.g. copy from arepository to the local storage)

CPU Activity

I/O Activity

Execution time

...

...

VM Image creationCloudlet input file

creation

CloudletCPU

utilization

I/O related CPU

utilization

I/O execution

Time

Figure – Time model after the extension

23/6/2016 11

Power Computation

Before the Extension

CloudSim power model : uses only hosts CPU utilization todetermine their power consumption

150

175

200

225

250

275

300

0 60 120 180 240 300 360 420 480 540 600 660 720

Idle mode power(0% CPU)

Operating mode power(CPU+RAM)

CPU Utilization

Po

wer

(w

att)

Time (second)

Power (watt) Pmax(100% CPU) Pidle(0% CPU)

Figure – Power model before the extension

23/6/2016 12

Power Computation

Before the Extension

CloudSim power model : uses only hosts CPU utilization todetermine their power consumption

150

175

200

225

250

275

300

0 60 120 180 240 300 360 420 480 540 600 660 720

Idle mode power(0% CPU)

Operating mode power(CPU+RAM)

CPU Utilization

Po

wer

(w

att)

Time (second)

Power (watt) Pmax(100% CPU) Pidle(0% CPU)

Figure – Power model before the extension

23/6/2016 12

Power Computation

Before the Extension

CloudSim power model : uses only hosts CPU utilization todetermine their power consumption

150

175

200

225

250

275

300

0 60 120 180 240 300 360 420 480 540 600 660 720

Idle mode power(0% CPU)

Operating mode power(CPU+RAM)

CPU Utilization

Po

wer

(w

att)

Time (second)

Power (watt) Pmax(100% CPU) Pidle(0% CPU)

Figure – Power model before the extension

23/6/2016 12

Power Computation

After the Extension

Our extension adds to the existing power model the powerconsumption generated by the I/O workload execution [2]

150

175

200

225

250

275

300

0 60 120 180 240 300 360 420 480 540 600 660 720 780 840

Idle mode power(0% CPU+Storage)

Operating mode power(CPU+RAM+Storage)

I/O Activity

CPU Utilization

Po

wer

(w

att)

Time (second)

Power (watt)Pmax(100% CPU)

Pidle(0% CPU)

Pmax(with storage)

Pidle(with storage)

Figure – Power model after the extension

[2] H. Ouarnoughi, J. Boukhobza, F. Singhoff, S. Rubini : ”A Cost Model for Virtual Machine Storage in Cloud IaaSContext”

23/6/2016 13

Power Computation

After the Extension

Our extension adds to the existing power model the powerconsumption generated by the I/O workload execution [2]

150

175

200

225

250

275

300

0 60 120 180 240 300 360 420 480 540 600 660 720 780 840

Idle mode power(0% CPU+Storage)

Operating mode power(CPU+RAM+Storage)

I/O Activity

CPU Utilization

Po

wer

(w

att)

Time (second)

Power (watt)Pmax(100% CPU)

Pidle(0% CPU)

Pmax(with storage)

Pidle(with storage)

Figure – Power model after the extension

[2] H. Ouarnoughi, J. Boukhobza, F. Singhoff, S. Rubini : ”A Cost Model for Virtual Machine Storage in Cloud IaaSContext”

23/6/2016 13

Power Computation

After the Extension

Our extension adds to the existing power model the powerconsumption generated by the I/O workload execution [2]

150

175

200

225

250

275

300

0 60 120 180 240 300 360 420 480 540 600 660 720 780 840

Idle mode power(0% CPU+Storage)

Operating mode power(CPU+RAM+Storage)

I/O Activity

CPU Utilization

Po

wer

(w

att)

Time (second)

Power (watt)Pmax(100% CPU)

Pidle(0% CPU)

Pmax(with storage)

Pidle(with storage)

Figure – Power model after the extension

[2] H. Ouarnoughi, J. Boukhobza, F. Singhoff, S. Rubini : ”A Cost Model for Virtual Machine Storage in Cloud IaaSContext”

23/6/2016 13

Storage Device Support

Before the Extension

Native CloudSim storage system :an HDD-based SAN,shared with all data center’s hosts

Storage system performances :f(max(HDDP erformance, SANT hroughput))

HDD

SAN Storage

HDD HDD

...

...

Host

...

...Host

Figure – Data center storage system before the extension

23/6/2016 14

Storage Device Support

Before the Extension

Native CloudSim storage system :an HDD-based SAN,shared with all data center’s hosts

Storage system performances :f(max(HDDP erformance, SANT hroughput))

HDD

SAN Storage

HDD HDD

...

...

Host

...

...Host

Figure – Data center storage system before the extension

23/6/2016 14

Storage Device Support

Before the Extension

Native CloudSim storage system :an HDD-based SAN,shared with all data center’s hosts

Storage system performances :f(max(HDDP erformance, SANT hroughput))

HDD

SAN Storage

HDD HDD

...

...

Host

...

...Host

Figure – Data center storage system before the extension

23/6/2016 14

Storage Device Support

Before the Extension

Native CloudSim storage system :an HDD-based SAN,shared with all data center’s hosts

Storage system performances :f(max(HDDP erformance, SANT hroughput))

HDD

SAN Storage

HDD HDD

...

...

Host

...

...Host

Figure – Data center storage system before the extension

23/6/2016 14

Storage Device Support

After the Extension

Three main storage system capabilities :

Support for flash-based SSD

Support for local storage (i.e. DAS storage)

More storage device attributes (e.g. power, performance,reliability, etc)

...

Host

...

...Host

SAN Storage

HDD

...SSD HDD SSD

DAS Storage

HDD

...SSD

DAS Storage

HDD

...SSD

Figure – Data center storage system after the extension

23/6/2016 15

Storage Device Support

After the Extension

Three main storage system capabilities :

Support for flash-based SSD

Support for local storage (i.e. DAS storage)

More storage device attributes (e.g. power, performance,reliability, etc)

...

Host

...

...Host

SAN Storage

HDD

...SSD HDD SSD

DAS Storage

HDD

...SSD

DAS Storage

HDD

...SSD

Figure – Data center storage system after the extension

23/6/2016 15

Storage Device Support

After the Extension

Three main storage system capabilities :

Support for flash-based SSD

Support for local storage (i.e. DAS storage)

More storage device attributes (e.g. power, performance,reliability, etc)

...

Host

...

...Host

SAN Storage

HDD

...SSD HDD SSD

DAS Storage

HDD

...SSD

DAS Storage

HDD

...SSD

Figure – Data center storage system after the extension

23/6/2016 15

Storage Device Support

After the Extension

Three main storage system capabilities :

Support for flash-based SSD

Support for local storage (i.e. DAS storage)

More storage device attributes (e.g. power, performance,reliability, etc)

...

Host

...

...Host

SAN Storage

HDD

...SSD HDD SSD

DAS Storage

HDD

...SSD

DAS Storage

HDD

...SSD

Figure – Data center storage system after the extension

23/6/2016 15

Storage Device Support

After the Extension

Three main storage system capabilities :

Support for flash-based SSD

Support for local storage (i.e. DAS storage)

More storage device attributes (e.g. power, performance,reliability, etc)

...

Host

...

...Host

SAN Storage

HDD

...SSD HDD SSD

DAS Storage

HDD

...SSD

DAS Storage

HDD

...SSD

Figure – Data center storage system after the extension

23/6/2016 15

Storage Device Support

After the Extension

Three main storage system capabilities :

Support for flash-based SSD

Support for local storage (i.e. DAS storage)

More storage device attributes (e.g. power, performance,reliability, etc)

...

Host

...

...Host

SAN Storage

HDD

...SSD HDD SSD

DAS Storage

HDD

...SSD

DAS Storage

HDD

...SSD

Figure – Data center storage system after the extension

23/6/2016 15

I/O Workload and CPU Time

CPU time for I/O Workload

VM I/O workload execution : storage system activities +CPU utilization due to I/O software stack execution

I/O workload → CPU utilization : a correlation model thatgives CPU time depending on I/O workload characteristics

0 %

10 %

20 %

30 %

40 %

50 %

4 8 16 32 64 128

CP

U T

ime

IO Size (KB)

CPU Time for Read I/O Execution

HDD random readHDD sequential read

SSD random readSSD sequential read

0 %0.5 %

1 %1.5 %

2 %2.5 %

3 %3.5 %

4 8 16 32 64 128

CP

U T

ime

I/O Size (KB)

CPU Time for Write I/O Execution

HDD random writeHDD sequential write

SSD random writeSSD sequential write

23/6/2016 16

I/O Workload and CPU Time

CPU time for I/O Workload

VM I/O workload execution : storage system activities +CPU utilization due to I/O software stack execution

I/O workload → CPU utilization : a correlation model thatgives CPU time depending on I/O workload characteristics

0 %

10 %

20 %

30 %

40 %

50 %

4 8 16 32 64 128

CP

U T

ime

IO Size (KB)

CPU Time for Read I/O Execution

HDD random readHDD sequential read

SSD random readSSD sequential read

0 %0.5 %

1 %1.5 %

2 %2.5 %

3 %3.5 %

4 8 16 32 64 128

CP

U T

ime

I/O Size (KB)

CPU Time for Write I/O Execution

HDD random writeHDD sequential write

SSD random writeSSD sequential write

23/6/2016 16

I/O Workload and CPU Time

CPU time for I/O Workload

VM I/O workload execution : storage system activities +CPU utilization due to I/O software stack execution

I/O workload → CPU utilization : a correlation model thatgives CPU time depending on I/O workload characteristics

0 %

10 %

20 %

30 %

40 %

50 %

4 8 16 32 64 128

CP

U T

ime

IO Size (KB)

CPU Time for Read I/O Execution

HDD random readHDD sequential read

SSD random readSSD sequential read

0 %0.5 %

1 %1.5 %

2 %2.5 %

3 %3.5 %

4 8 16 32 64 128

CP

U T

ime

I/O Size (KB)

CPU Time for Write I/O Execution

HDD random writeHDD sequential write

SSD random writeSSD sequential write

23/6/2016 16

I/O Workload and CPU Time

CPU time for I/O Workload

VM I/O workload execution : storage system activities +CPU utilization due to I/O software stack execution

I/O workload → CPU utilization : a correlation model thatgives CPU time depending on I/O workload characteristics

0 %

10 %

20 %

30 %

40 %

50 %

4 8 16 32 64 128

CP

U T

ime

IO Size (KB)

CPU Time for Read I/O Execution

HDD random readHDD sequential read

SSD random readSSD sequential read

0 %0.5 %

1 %1.5 %

2 %2.5 %

3 %3.5 %

4 8 16 32 64 128

CP

U T

ime

I/O Size (KB)

CPU Time for Write I/O Execution

HDD random writeHDD sequential write

SSD random writeSSD sequential write

23/6/2016 16

Evaluation

23/6/2016 17

Evaluation : Summary

Evaluation Methodology

Two steps to evaluate CloudSim storage extensions :

Real workload execution : running experiments in a realenvironment + collecting CPU, memory, and I/O measures.

Workload simulation : replaying the same scenario inCloudSim + turning the storage system ON/OFF

SSD HDDStorage System

VM1 VM2 VMn

ffmpeg Video Transcoding

Host Operating System Input/Output Parameters

Validation Tools

Hardware/Software Platform

Traces Collector(CPU, RAM, I/O)

KVM/Qemu Hypervisor

Real Traces● CPU Utilization● RAM Utilization● I/O Traces

CloudSim

Simulation Results● Execution Time● Energy Consumption

Simulation Results● Execution Time● Energy Consumption

Storage System

OFF

Storage System

ON

Storage System Impact ● Energy results comparison

1

2

3

4

Figure – Evaluation methodology

23/6/2016 18

Evaluation : Summary

Evaluation Methodology

Two steps to evaluate CloudSim storage extensions :

Real workload execution : running experiments in a realenvironment + collecting CPU, memory, and I/O measures.

Workload simulation : replaying the same scenario inCloudSim + turning the storage system ON/OFF

SSD HDDStorage System

VM1 VM2 VMn

ffmpeg Video Transcoding

Host Operating System Input/Output Parameters

Validation Tools

Hardware/Software Platform

Traces Collector(CPU, RAM, I/O)

KVM/Qemu Hypervisor

Real Traces● CPU Utilization● RAM Utilization● I/O Traces

CloudSim

Simulation Results● Execution Time● Energy Consumption

Simulation Results● Execution Time● Energy Consumption

Storage System

OFF

Storage System

ON

Storage System Impact ● Energy results comparison

1

2

3

4

Figure – Evaluation methodology

23/6/2016 18

Evaluation : Summary

Evaluation Methodology

Two steps to evaluate CloudSim storage extensions :

Real workload execution : running experiments in a realenvironment + collecting CPU, memory, and I/O measures.

Workload simulation : replaying the same scenario inCloudSim + turning the storage system ON/OFF

SSD HDDStorage System

VM1 VM2 VMn

ffmpeg Video Transcoding

Host Operating System Input/Output Parameters

Validation Tools

Hardware/Software Platform

Traces Collector(CPU, RAM, I/O)

KVM/Qemu Hypervisor

Real Traces● CPU Utilization● RAM Utilization● I/O Traces

CloudSim

Simulation Results● Execution Time● Energy Consumption

Simulation Results● Execution Time● Energy Consumption

Storage System

OFF

Storage System

ON

Storage System Impact ● Energy results comparison

1

2

3

4

Figure – Evaluation methodology

23/6/2016 18

Evaluation : Summary

Evaluation Methodology

Two steps to evaluate CloudSim storage extensions :

Real workload execution : running experiments in a realenvironment + collecting CPU, memory, and I/O measures.

Workload simulation : replaying the same scenario inCloudSim + turning the storage system ON/OFF

SSD HDDStorage System

VM1 VM2 VMn

ffmpeg Video Transcoding

Host Operating System Input/Output Parameters

Validation Tools

Hardware/Software Platform

Traces Collector(CPU, RAM, I/O)

KVM/Qemu Hypervisor

Real Traces● CPU Utilization● RAM Utilization● I/O Traces

CloudSim

Simulation Results● Execution Time● Energy Consumption

Simulation Results● Execution Time● Energy Consumption

Storage System

OFF

Storage System

ON

Storage System Impact ● Energy results comparison

1

2

3

4

Figure – Evaluation methodology

23/6/2016 18

Evaluation : Summary

Evaluation Methodology

Two steps to evaluate CloudSim storage extensions :

Real workload execution : running experiments in a realenvironment + collecting CPU, memory, and I/O measures.

Workload simulation : replaying the same scenario inCloudSim + turning the storage system ON/OFF

SSD HDDStorage System

VM1 VM2 VMn

ffmpeg Video Transcoding

Host Operating System Input/Output Parameters

Validation Tools

Hardware/Software Platform

Traces Collector(CPU, RAM, I/O)

KVM/Qemu Hypervisor

Real Traces● CPU Utilization● RAM Utilization● I/O Traces

CloudSim

Simulation Results● Execution Time● Energy Consumption

Simulation Results● Execution Time● Energy Consumption

Storage System

OFF

Storage System

ON

Storage System Impact ● Energy results comparison

1

2

3

4

Figure – Evaluation methodology23/6/2016 18

Workload execution

Real workload execution

The real workload execution phaseincludes two steps :

Workload execution : VMsrunning encoding videobenchmark as a real use case

Trace collection : monitoringthe host CPU, memoryutilization + tracing I/Os SSD HDD

Storage System

VM1 VM2 VMn

ffmpeg Video Transcoding

Host Operating System

Traces Collector(CPU, RAM, I/O)

KVM/Qemu Hypervisor

1

2

23/6/2016 19

Workload execution

Real workload execution

The real workload execution phaseincludes two steps :

Workload execution : VMsrunning encoding videobenchmark as a real use case

Trace collection : monitoringthe host CPU, memoryutilization + tracing I/Os SSD HDD

Storage System

VM1 VM2 VMn

ffmpeg Video Transcoding

Host Operating System

Traces Collector(CPU, RAM, I/O)

KVM/Qemu Hypervisor

1

2

23/6/2016 19

Workload execution

Real workload execution

The real workload execution phaseincludes two steps :

Workload execution : VMsrunning encoding videobenchmark as a real use case

Trace collection : monitoringthe host CPU, memoryutilization + tracing I/Os

SSD HDDStorage System

VM1 VM2 VMn

ffmpeg Video Transcoding

Host Operating System

Traces Collector(CPU, RAM, I/O)

KVM/Qemu Hypervisor

1

2

23/6/2016 19

Workload execution

Real workload execution

The real workload execution phaseincludes two steps :

Workload execution : VMsrunning encoding videobenchmark as a real use case

Trace collection : monitoringthe host CPU, memoryutilization + tracing I/Os SSD HDD

Storage System

VM1 VM2 VMn

ffmpeg Video Transcoding

Host Operating System

Traces Collector(CPU, RAM, I/O)

KVM/Qemu Hypervisor

1

2

23/6/2016 19

Workload simulation

Real workload execution

The workload simulation phaseincludes two steps :

Simulation using real traces :replaying traces in CloudSim +varying parameters (i.e. numberof VMs/Host, storage systemON/OFF, and storage deviceHDD/SSD)

Simulation results comparison :power consumption to quantifyI/O workload execution impact

Real Traces● CPU Utilization● RAM Utilization● I/O Traces

CloudSim

Simulation Results● Execution Time● Energy Consumption

Storage System

OFF

Storage System

ON

Storage System Impact ● Energy results comparison

1

2

Simulation Results● Execution Time● Energy Consumption

23/6/2016 20

Workload simulation

Real workload execution

The workload simulation phaseincludes two steps :

Simulation using real traces :replaying traces in CloudSim +varying parameters (i.e. numberof VMs/Host, storage systemON/OFF, and storage deviceHDD/SSD)

Simulation results comparison :power consumption to quantifyI/O workload execution impact

Real Traces● CPU Utilization● RAM Utilization● I/O Traces

CloudSim

Simulation Results● Execution Time● Energy Consumption

Storage System

OFF

Storage System

ON

Storage System Impact ● Energy results comparison

1

2

Simulation Results● Execution Time● Energy Consumption

23/6/2016 20

Workload simulation

Real workload execution

The workload simulation phaseincludes two steps :

Simulation using real traces :replaying traces in CloudSim +varying parameters (i.e. numberof VMs/Host, storage systemON/OFF, and storage deviceHDD/SSD)

Simulation results comparison :power consumption to quantifyI/O workload execution impact

Real Traces● CPU Utilization● RAM Utilization● I/O Traces

CloudSim

Simulation Results● Execution Time● Energy Consumption

Storage System

OFF

Storage System

ON

Storage System Impact ● Energy results comparison

1

2

Simulation Results● Execution Time● Energy Consumption

23/6/2016 20

Workload simulation

Real workload execution

The workload simulation phaseincludes two steps :

Simulation using real traces :replaying traces in CloudSim +varying parameters (i.e. numberof VMs/Host, storage systemON/OFF, and storage deviceHDD/SSD)

Simulation results comparison :power consumption to quantifyI/O workload execution impact

Real Traces● CPU Utilization● RAM Utilization● I/O Traces

CloudSim

Simulation Results● Execution Time● Energy Consumption

Storage System

OFF

Storage System

ON

Storage System Impact ● Energy results comparison

1

2

Simulation Results● Execution Time● Energy Consumption

23/6/2016 20

Evaluation Metrics

Evaluation Metrics

Our evaluation results show two evaluation metrics :

Power consumption related to I/O processing → validatethe impact of I/Os on the overall power consumption

Difference between simulations using real traces and usingcorrelation model → validate its accuracy

23/6/2016 21

Evaluation Metrics

Evaluation Metrics

Our evaluation results show two evaluation metrics :

Power consumption related to I/O processing → validatethe impact of I/Os on the overall power consumption

Difference between simulations using real traces and usingcorrelation model → validate its accuracy

23/6/2016 21

Evaluation Metrics

Evaluation Metrics

Our evaluation results show two evaluation metrics :

Power consumption related to I/O processing → validatethe impact of I/Os on the overall power consumption

Difference between simulations using real traces and usingcorrelation model → validate its accuracy

23/6/2016 21

Evaluation Results & Discussion

0 %

5 %

10 %

15 %

20 %

2 4 6 8

I/O P

roce

ssin

g Re

late

d En

ergy

Con

sum

ptio

n

Number of VMs per Host

Storage System Energy: All Traces vs Correlation Model

HDD with all traces

10.52

15.9516.89 16.28

HDD with correlation model

8.43

12.96

15.89 15.16

SSD with all traces

2.13 2.89 3.37 3.09

SSD with correlation model

3.15 3.13 3.09 2.82

Discussion

Energy consumption ↗ when to the number of VMs/Host ↗Difference between HDD and SSD energy consumption

Correlation model vs real traces : ∼12% average error rate

23/6/2016 22

Evaluation Results & Discussion

0 %

5 %

10 %

15 %

20 %

2 4 6 8

I/O P

roce

ssin

g Re

late

d En

ergy

Con

sum

ptio

n

Number of VMs per Host

Storage System Energy: All Traces vs Correlation Model

HDD with all traces

10.52

15.9516.89 16.28

HDD with correlation model

8.43

12.96

15.89 15.16

SSD with all traces

2.13 2.89 3.37 3.09

SSD with correlation model

3.15 3.13 3.09 2.82

Discussion

Energy consumption ↗ when to the number of VMs/Host ↗Difference between HDD and SSD energy consumption

Correlation model vs real traces : ∼12% average error rate

23/6/2016 22

Evaluation Results & Discussion

0 %

5 %

10 %

15 %

20 %

2 4 6 8

I/O P

roce

ssin

g Re

late

d En

ergy

Con

sum

ptio

n

Number of VMs per Host

Storage System Energy: All Traces vs Correlation Model

HDD with all traces

10.52

15.9516.89 16.28

HDD with correlation model

8.43

12.96

15.89 15.16

SSD with all traces

2.13 2.89 3.37 3.09

SSD with correlation model

3.15 3.13 3.09 2.82

Discussion

Energy consumption ↗ when to the number of VMs/Host ↗

Difference between HDD and SSD energy consumption

Correlation model vs real traces : ∼12% average error rate

23/6/2016 22

Evaluation Results & Discussion

0 %

5 %

10 %

15 %

20 %

2 4 6 8

I/O P

roce

ssin

g Re

late

d En

ergy

Con

sum

ptio

n

Number of VMs per Host

Storage System Energy: All Traces vs Correlation Model

HDD with all traces

10.52

15.9516.89 16.28

HDD with correlation model

8.43

12.96

15.89 15.16

SSD with all traces

2.13 2.89 3.37 3.09

SSD with correlation model

3.15 3.13 3.09 2.82

Discussion

Energy consumption ↗ when to the number of VMs/Host ↗Difference between HDD and SSD energy consumption

Correlation model vs real traces : ∼12% average error rate

23/6/2016 22

Evaluation Results & Discussion

0 %

5 %

10 %

15 %

20 %

2 4 6 8

I/O P

roce

ssin

g Re

late

d En

ergy

Con

sum

ptio

n

Number of VMs per Host

Storage System Energy: All Traces vs Correlation Model

HDD with all traces

10.52

15.9516.89 16.28

HDD with correlation model

8.43

12.96

15.89 15.16

SSD with all traces

2.13 2.89 3.37 3.09

SSD with correlation model

3.15 3.13 3.09 2.82

Discussion

Energy consumption ↗ when to the number of VMs/Host ↗Difference between HDD and SSD energy consumption

Correlation model vs real traces : ∼12% average error rate

23/6/2016 22

Conclusion & Future Work

23/6/2016 23

Conclusion & Future Work

Conclusion

We propose an extension of CloudSim to consider VM I/Oworkload processing in form of three contributions :

Updating the time and energy models of CloudSim

Considering different storage devices (i.e. HDD and SSD)

Including a CPU correlation model that depends on I/Oworkload characteristics and storage device type

Future Work

Using this implementation in a VM placementoptimization approach

Studying and integrating the problem of interferencebetween VMs sharing the same storage device

23/6/2016 24

Conclusion & Future Work

Conclusion

We propose an extension of CloudSim to consider VM I/Oworkload processing in form of three contributions :

Updating the time and energy models of CloudSim

Considering different storage devices (i.e. HDD and SSD)

Including a CPU correlation model that depends on I/Oworkload characteristics and storage device type

Future Work

Using this implementation in a VM placementoptimization approach

Studying and integrating the problem of interferencebetween VMs sharing the same storage device

23/6/2016 24

Conclusion & Future Work

Conclusion

We propose an extension of CloudSim to consider VM I/Oworkload processing in form of three contributions :

Updating the time and energy models of CloudSim

Considering different storage devices (i.e. HDD and SSD)

Including a CPU correlation model that depends on I/Oworkload characteristics and storage device type

Future Work

Using this implementation in a VM placementoptimization approach

Studying and integrating the problem of interferencebetween VMs sharing the same storage device

23/6/2016 24

Conclusion & Future Work

Conclusion

We propose an extension of CloudSim to consider VM I/Oworkload processing in form of three contributions :

Updating the time and energy models of CloudSim

Considering different storage devices (i.e. HDD and SSD)

Including a CPU correlation model that depends on I/Oworkload characteristics and storage device type

Future Work

Using this implementation in a VM placementoptimization approach

Studying and integrating the problem of interferencebetween VMs sharing the same storage device

23/6/2016 24

Conclusion & Future Work

Conclusion

We propose an extension of CloudSim to consider VM I/Oworkload processing in form of three contributions :

Updating the time and energy models of CloudSim

Considering different storage devices (i.e. HDD and SSD)

Including a CPU correlation model that depends on I/Oworkload characteristics and storage device type

Future Work

Using this implementation in a VM placementoptimization approach

Studying and integrating the problem of interferencebetween VMs sharing the same storage device

23/6/2016 24

Conclusion & Future Work

Conclusion

We propose an extension of CloudSim to consider VM I/Oworkload processing in form of three contributions :

Updating the time and energy models of CloudSim

Considering different storage devices (i.e. HDD and SSD)

Including a CPU correlation model that depends on I/Oworkload characteristics and storage device type

Future Work

Using this implementation in a VM placementoptimization approach

Studying and integrating the problem of interferencebetween VMs sharing the same storage device

23/6/2016 24

Conclusion & Future Work

Conclusion

We propose an extension of CloudSim to consider VM I/Oworkload processing in form of three contributions :

Updating the time and energy models of CloudSim

Considering different storage devices (i.e. HDD and SSD)

Including a CPU correlation model that depends on I/Oworkload characteristics and storage device type

Future Work

Using this implementation in a VM placementoptimization approach

Studying and integrating the problem of interferencebetween VMs sharing the same storage device

23/6/2016 24

Conclusion & Future Work

Conclusion

We propose an extension of CloudSim to consider VM I/Oworkload processing in form of three contributions :

Updating the time and energy models of CloudSim

Considering different storage devices (i.e. HDD and SSD)

Including a CPU correlation model that depends on I/Oworkload characteristics and storage device type

Future Work

Using this implementation in a VM placementoptimization approach

Studying and integrating the problem of interferencebetween VMs sharing the same storage device

23/6/2016 24

My wish...

More research experience, so Iam looking for a postdoctoral

position !

23/6/2016 25

Thanks

{ hamza.ouarnoughi@b-com.com }

www.b-com.com