EFFICIENT LIVE VM MIGRATION IN A CLOUD

Pabba Praneeth N Yaswanth P Rakesh B100864CS B100976CS B100565CS Y Sreenivasa Reddy B100825CS

Guided by : Mr. Saidalavi Kalady

Problem Statement

To simulate a federated cloud environment with dynamic load balancing by VM migration and simulation of real time cloud market scenario.

Federated Cloud

A Federated cloud is an interconnected group of diverse interacting clouds with regulated configuration and features. Clouds coordinate with each other to offer effective and seamless QoS levels in different geographic locations. Sharply increased demand and demand at unreachable geographic frontiers could be effectively managed with coordination.

VM Migration

Live VM Migration is the process of migrating a VM instance as one unit from one physical host to another host while continuously running it. Advantages:

Load balancing

Fault management
Energy cost savings in cloud Managing regionally distributed demand

CloudSim

CloudSim is a generalized and extensible Cloud Computing model simulator Primary uses:

For benchmark experiments in a repeatable, dependable and scalable environment Tuning system bottlenecks before deploying in real time clouds Various resource provisioning and application models tested with ease Experiment failures only crash simulation causing no real world damages

CloudSim Features and limitations

Features:

Simulation of large cloud computing environments on a single physical node Supports modelling of real world entities such as datacenters, brokers etc.

Cloud federation, SLA management and other core competencies can be modelled and simulated effectively
Real world topology and network delays in cloud induced using BRITE topology Deploying real applications is NOT possible in CloudSim as it only a simulator and doesnt execute real applications on cloud.

Limitation:

CloudSim Architecture

Proposed Design
End User

Resource Broker Agent

Resource Provider Agent

Resource Provider Agent

Resource Provider Agent

Cloud Coordinator Cloud 1

Cloud Coordinator Cloud 2

Cloud Coordinator Cloud 3

Analogous Terms and Definitions

In CloudSim DataCenter DataCenterBroker Cloudlet In Real Time Cloud Resource Broker Agent Application Service to be deployed Definition Virtualization of resources Agent between user and the federated cloud Task assigned to a cloud which has to be completed by a VM

FederatedDataCenter
Host EndUser (added)

A cloud in federation
Physical host or server in cloud End user wishing to deploy VMs in a cloud

<NA>
Application or Data Server in cloud <NA>

Design Implementation
End User EndUser DataCenter Broker

Resource Broker Agent

Resource Provider Agent

Resource Provider Agent

Resource Provider Agent

CloudCoordi nator

Cloud Coordinator Cloud 1

Cloud Coordinator Cloud 2

Cloud Coordinator Cloud 3

Federated DataCenter

Implementation

Implementation: Extended Class Diagram

Extended Class Diagram Red classes added and blue classes modified in source code

Addition of End User component

End User component is added to the simulation specification layer in CloudSim architecture. Application requests (Cloudlets) and VM configuration form the basic requirement on end user side. They are sent from EndUser to DataCenterBroker who packages all such requests and deploys them on best fit clouds. FURTHER SCOPE: Billing to respective users based on usage and SLA management could be implemented by enhancing EndUser component. Cloud Scenario

End User

User Requirements

Application Configuration

User or DataCenterBroker

Extension of VmAllocationPolicy

VmAllocationPolicy is an abstract class which is responsible for allocating VMs to different physical hosts in a cloud based on VM and Host configurations.
VmAllocationPolicy being abstract cannot be directly implemented and is extended by default VmAllocationPolicySimple of CloudSim. VmAllocationPolicySimple allocates VM to host with least Pes (Processing entities) in use. Our project extends VmAllocationPolicy to allocate VMs based on RAV (Resource Availability Vector) of a host.

Resource Availability Vector

Resource fraction vector is a vector with magnitude equal to ratio of available resource to the total available resource on a host. Every resource fraction vector is aligned in a dedicated direction. Resource Availability Vector is the linear combination of all Resource fraction vectors of the host. =( )+( )+() Here, Resource fraction vectors belong to bandwidth, RAM and Processing power respectively. All resources are taken at equal weightage in linear combination.

The host with maximum RAV magnitude gets the VM assigned.

( ) +( ) + ()

Why RAV???

The idea of availability vectors is used throughout the project.

RAV represents the utilization and availability of all resources without neglecting any resource used by a host. Hence, scheduling decisions involving RAV tend to provide optimal performance metrics. In case any resource needs greater importance, its weight in linear combination could be increased.

Extension of CloudCoordinator

CloudCoordinator is the root component for a cloud in federation.

It maintains information about other clouds in the federation and forms a medium of communication for information and data interchange between clouds. updateDataCenter() function polls all the resources of the cloud regularly to check if any host is overloaded or lightly loaded and is a target for VM migration. migrate() function implements migration policy of VM from one cloud to another.

migrate() in CloudCoordinator

migrate() already obtains overloaded Host from the sensors of the cloud it triggers. In order to fix the target host for migration, it uses utilization vector similar to RAV as follows: =( ) + ( ) + ( ) + ( )

Here, an extra flag In Same DataCenter is added to check if host under consideration in within the same cloud. Flag is assigned 0 if host is within same cloud and 1 otherwise. This decision is taken based on overweighing migration overhead costs and exchange costs for migration of VM between different clouds in federation. Decision policy is similar to Child is better off with mother rather than a caretaker.

Final step of migrate()

Once target host is set in migrate() , we need to find the VM from overloaded to target host for migration.

The most heavyweight VM is found and migrated to restore balance in load as follows:
=()+()+( ) The final step of migrate() is performed by invoking send() or sendNow() function of SimEntity core class of CloudSim.

Creation of Sensors

A sensor is a entity that responds to a certain input from a physical or intellectual environment.

Sensors are used in clouds to detect abnormal resource usages in clouds and respond accordingly.
Sensor is implemented as an interface in CloudSim. Our work extends Sensor to create 3 sensors : BWSensor, RAMSensor and ProcessSensor to monitor bandwidth, RAM and Processing power respectively.

Working and use of Sensors

Sensors have lower and upper thresholds as set values.

monitor() function of a sensor monitors utilization of all physical components if usage moves beyond range specified by thresholds.
In case of an abnormality, sensor reports it appropriate return value: +1 overutilization and -1 for underutilization. monitor() of all sensors is invoked by updateDataCenter() of each cloud regularly.

Working model of Implementation

Cloud 1

Cloud 2