First National Workshop of Cloud ComputingAmirkabir University of Technology

Persented by: Neda Malekinedamaleki87@gmail.com

CloudSim: A Toolkit for Modeling and Simulation of

Cloud Computing Environments

OutLine

• Introduction• Related Work• CloudSim Architecture• CloudSim Modelings• Design and Implementation• CloudSim Steps• Conclusions and Future works• Green Cloud

Introduction(1/2):Cloud

• Cloud computing delivers: XaaS

• X :{Software, Platform, Infrastructure }

So users can access and deploy applications from anywhere in the Internet driven by demand and QoS requirements.

Introduction(2/2):Why Simulation?

Cloud Providor Challenges:•Maintain Quality of Service•Efficient Resourse Utilization•Dynamic Workload•Violation of Service Level Agreement•Difficulties in Testing

It’s not possible to perform benchmarking experiments in repeatable, dependable, and

scalable environment using real-world Cloud.

Possible alternative: Simulation Tool

Related Works

Grid simulators: GridSim SimGrid OptoSim GangSim

But none of them are able to isolate the multi-layer service abstractions(SaaS/PaaS/IaaS) differentiation and model the virtualized resources required by Cloud. So:

Main Contribution: CloudSim

A holistic software framework for modeling Cloud computing environments

And Performance testing application services.

Features & AdvantagesFeatures• Discrete Time Event-Driven• Support modeling and simulation of large scale

Cloud computing environments, including data centers

• Support simulation of network connections among simulated elements

Advantages• Time effectiveness• Flexibility and applicability• Test policies in repeatable and controllable

environment• Tune system bottlenecks before deploying on real

clouds

Layered CloudSim Architecture(1/7)

Modeling in Cloudsim (1/5)

Modeling DataCenter Modeling VM Allocation Modeling Network Behavior Modeling Dynamic Workloads Modeling Power Consumption

CloudSim Steps(1/2)

a

broker

(VMs , Apps)

Cloud Information Service(CIS)

Is Registered all Datacenters and

their characteristics

Cloud Datacenter A

Cloud Datacenter B

Cloud Datacenter C

Query

Availa

ble D

atac

ente

rs

Allocation

Allocation Policies: Enough Capacity,Ram,Storage,Bandwid

th

VM1,V10,VM6,VM7

VM2,VM4

VM9,V3,VM5

VM8

Scheduling Policies: Sharing of Host Mips between VMs

• Space Shared•Time Shared

DataCenter Modeling Number of Hosts, VMs and Cloudlets (tasks) o Host(mips, ram, storage, bandwidth)o Datacenter(arch, os, vmm, hostlist, cost

mem/bw/storage)

VMo MIPS, pesNumber(no. of cpu), Ram(MB),

BW(MB/s)

Cloudleto Length (MI), pesNumber, input Size, output Size

VM Allocation Modeling• Time Shared policy• Space Shared Policy

Simulation Setup:

========== OUTPUT ==========Cloudlet ID STATUS Data center ID VM ID Time Start Time Finish

Time0 SUCCESS 2 0 2 0.1

2.12 SUCCESS 2 0 2 0.1

2.11 SUCCESS 2 1 2 0.1

2.13 SUCCESS 2 1 2 0.1

2.1*****Datacenter: Datacenter_0*****

User id Debt 3 2051.2

1 datecenter 1 dual-core host, each core'mips: 1000 2 vm, mips:1000 4 cloudlets, length: 1000mips core1 deal with two cloudlets(t1 and t2), and core2

deal with the other two cloudlets(t3 and t4), so, all cloudlets should finished at 2.1s

Network Modeling

• Latency Matrix

Delay time from entity i to entity j

Entity i Entity j

Dynamic Workload Modeling• The Strategy is to Vary VM

Utilization!

25% 43% 60% 30% 10% 90% ….

Delay= not all the time, CPU is

utilized

Design and Implementation(1/2)

CloudSim Class Design Diagram

Design and Implementation(2/2)Simulation Data Flow

Design and Impelementation(3/4) CloudSim Sequence Diagram

Conclusion

Time effectiveness

Flexibility and applicability

Test services in repeatable and controllable

environment

Tune system bottlenecks before deploying on

real clouds

Experiment with different workload mix

Green Cloud

Power(1/4):Powering Cloud Infrastructure

• Modern data centers, operating under the Cloud computing model, are hosting a variety of applications ranging from those that run for a few seconds (e.g. serving requests of web applications such as e-commerce and social networks portals) to those that run for longer periods of time (e.g. large dataset processing).

• So, Cloud Data Centers consume excessive amount of energy:• According to McKinsey report on “Revolutionizing Data

Center Energy Efficiency” :• A typical data center consumes as much energy as

25,000 households!!!• The total energy bill for data centers in 2010 was over

$11 billion and energy costs in a typical data center doubles every five years.

Power (1/2) Data centers are not only

expensive to maintain, but also unfriendly to the environment.

High energy costs and huge carbon emission are incurred due to the massive amount of electricity needed to power and cool the numerous servers hosted in these data centers.

Power Consumption in the Datacenter

Compute resources and particularly servers are at the heart of a complex, evolving system! They Consumes most power.

Where Does the Go?

Google Datacenter 2007

Power

Levels of Power Consideration(1/2):

System level

The objective of PA computing/communications is to improve power management and consumption using the awareness of power consumption of devices.

Recent devices (CPU, disk, communication links, etc.) support multiple power modes.

DVS(Dynamic Voltage Scaling)• DVS (Dynamic Voltage Scaling) technique

– Reducing the dynamic energy consumption by lowering the supply voltage at the cost of performance degradation

– Recent processors support such ability to adjust the supply voltage dynamically.– The dynamic energy consumption = * Vdd2 * f

• Vdd : the supply voltage• f : the number of clock cycle

• An example

5.02

10ms 25ms

deadlinepower

power deadline

10ms 25ms

(a) Supply voltage = 5.0 V (b) Supply voltage = 2.0 V

2.02

Levels of Power Consideration(2/2): DataCenter Level

Data center level

Virtualization

System resources

Target systems

Goal

Power saving techniques

Workload

Yes

No

Multiple resources

Single resource

Homogeneous

Heterogeneous

Minimize power / energy consumption

Minimize performance loss

DVFS

Meet power budget

Resource throttling

DCD

Arbitrary

Real-time applications

HPC-applications

Workload consolidation

A Key to Power Saving!

Power On Power Off

Pool of physical computer

nodes

Virtualization layer (VMMs, local resources managers)

Consumer, scientific and business applications

Global resource managers

User User User

VM provisioning SLA negotiation Application requests

Virtual Machines

andusers’

applications

WWW: Three Sub Problems

• When to migrate VMs?• Host overload detection algorithms• Host underload detection algorithms

• Which VMs to migrate?• VM selection algorithms

• Where to migrate VMs?• VM placement algorithms

Algorithms in each w Host overload detection

Adaptive utilization threshold based algorithmsMedian Absolute Deviation algorithm (MAD) Interquartile Range algorithm (IQR)

Regression based algorithms• Local Regression algorithm (LR)• Robust Local Regression algorithm (LRR)

Host underload detection algorithms Migrating the VMs from the least utilized host

VM selection algorithms Minimum Migration Time policy (MMT) Random Selection policy (RS) Maximum Correlation policy (MC)

VM placement algorithms Heuristic for the bin-packing problem – Power-Aware

Best Fit Decreasing algorithm (PABFD)

Performance MetricsSLA violation metrics

• Overloading Time Fraction (OTF) - the time fraction, during which active hosts experienced the 100% CPU utilization

• Performance Degradation due to VM Migrations (PDM)

• A combined SLA Violation metric (SLAV): SLAV = OTF * PDM

A combined metric that captures both energy consumption and the level of SLA violations, Energy and SLA Violation (ESV): ESV = Energy * SLAV

Real Workloads• Workload traces from more than 1000 VMs from

servers located in more than 500 places around the world.

• The data were obtained from the CoMon project, a monitoring infrastructure for PlanetLab

• PlanetLab is a distributed execution environment for doing benchmarked experiments . Totally it is a global research network that supports the development of new network services.

• A Data Center consisting 800 heterogeneous physical servers containing HP ProLiant ML110 G4 and HP ProLiant ML110 G5 servers.

• More than 1000 Heterogeneous VMs corresponding to Amazon EC2 instance types.

Content of WorkLoad Files These files contain CPU utilization values measured

every 5 minutes in PlanetLab's VMs for one day so:One day=24 hours= 5minutes*288

CloudSim contain a class called : UtilizationModelPlanetLabInMemory

which can be used to read those workload traces. An example: String inputFolder =

Dvfs.class.getClassLoader().getResource("workload/planetlab").getPath();

String outputFolder = "output"; String workload = "20110303"; // PlanetLab workload

Number of Samples

References R. Buyya, A. Beloglazov, J. Abawajy,

Energy-Efficient Management of Data Center Resources for Cloud Computing: A Vision, Architectural Elements, and Open Challenges, Proceedings of the 2010 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA2010), Las Vegas, USA, July 12-15, 2010.

A. Beloglazov, R. Buyya, Y. Lee, A. Zomaya, A Taxonomy and Survey of Energy-Efficient Data Centers and Cloud Computing Systems, Advances in Computers, Volume 82, 47-111pp, M. Zelkowitz (editor), Elsevier, Amsterdam, The Netherlands,March 2011.

S. Garg, C. Yeo, A Anandasivam, R. Buyya, Environment-Conscious Scheduling of HPC Applications on Distributed Cloud-oriented Data Centers, Journal of Parallel and Distributed Computing, 71(6):732-749, Elsevier Press, Amsterdam, The Netherlands, June 2011.

Thanks for your attention!

Any Questions , Suggestions and Comments?