Article Preview
Top1. Introduction
Cloud computing is currently most a promising approach to hosting data and applications globally shared resources. Cloud computing can also provide cost effective service in areas like IoT, Big Data and Embedded Systems (Bhatt et al., 2017; Dey and Mukherjee, 2016; Dey et al., 2018). Cloud service providers use data centers to provide computational resources as per user demand. Currently day by day many small e-businesses are shifted towards cloud and this increases number of data centers day by day.
Live migration of VMs from one PM to other can balance load and reduce power consumption of data centers. Live migration also plays main role to reduce performance overheads on PMs by performing resource redistribution amongst the PMs. VM live migration can perform consolidation of VMs to decrease the energy requirements for cloud data centers after shifting under-loaded PMs to sleep or power saving stage without decrease of QoS. So, multi-objective consolidation of VMs effectively provide improved PM resource utilization to reduce overall energy consumption and achieve “green cloud computing.”
Unfortunately, most of available approaches for VM live migration are narrow objective and selectively focusing on optimization of overheads associated with either pre or post migration phase. These types of partial optimization strategies focusing of either pre or post migration suffers from additional overheads on CPU, memory and network bandwidth availability with leads to increase in VM live migration span, application unavailability duration (downtime) and SLA violation. These narrow objectives like single aim to reduce requirement for active servers or decreasing n/w traffic for optimization strategies associated with VM live migration and consolidation makes these approaches unfit for practical implementation.
To obtain effective consolidation by using live migration multiple parameters and overheads must be considered. Any framework only focusing on few or limited objectives will result in biased solution for live migration and consolidation and will further adversely affect the VM performance in other aspects. However, live migration mostly results in some downtime for migrating VM which negatively affects services running on it and results in SLA violation. Also additional overheads are imposed on CPU, memory, network bandwidth of not only migrating VM but also source and destination PMs and their currently associated VMs, it is known as co-location interference. So, a live migration approach must be broad enough to focus on all these aspects.
1.1. Phases Involved During VM Live Migration
VM live migration involves following 6 phases:
- •
Target selection: VM is selected from a host for future migration.
- •
Destination Selection and Resource Reservation: A destination host is selected and required resources like CPU cores, partitioned memory and disk space are reserved on it.
- •
Repetitive Data Transfer: All the data in memory of VM is copied to destination host. Again pages which got modified during previous copy are being continuously copied to destination host.
- •
Suspension and Traffic Migration: VM on source host is suspended, all dirty pages of memory are sent finally and the notification is sent to hypervisor for migration all further incoming traffic of VM on source to corresponding destination.
- •
Resource De-Acquisition: Once notification is received from central cloud controller, all data associated with VM on source is copied to destination and all resources of VM associated with source host are released.
- •
Activation: VM on destination activated and all normal operations are started.