Enhancing minimal virtual machine migration in cloud environment

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
__________________________________________________________________________________________
Volume: 03 Special Issue: 07 | May-2014, Available @ http://www.ijret.org 375
ENHANCING MINIMAL VIRTUAL MACHINE MIGRATION IN CLOUD
ENVIRONMENT
Lidin Das1
, P Mohamed Shameem2
1
M.Tech Student, Dept. of CSE, TKM Institute of Technology, Kerala, India
2
Associate Professor, Dept. of CSE, TKM Institute of Technology, Kerala, India
Abstract
Virtualization is a popular solution that acts as a backbone for provisioning requirements of a cloud-based solution. And virtual
machine migration is key enabler for dynamic resource management in cloud-based systems. Live virtual machine migration transfers
the “state” of a virtual machine from one physical machine to another thereby can mitigate overloaded conditions and enables
uninterrupted maintenance activities. In this paper we will come across three main scenarios in virtual machine migration: when,
which and where to migrate. Main discussion area in this paper is the scenario, “where to migrate”, to choose the destination node to
which virtual machine get migrated. A bad choice would lead to a cascade in migration and thereby will create a cyclic effect. So we
have to select the better node in order to minimize further migration. For this, we propose a MVMM algorithm to minimize the virtual
machine migration.
Keywords: Cloud Computing, Hot Spot, Live Migration, Virtualization, Virtual Machine
-----------------------------------------------------------------------***----------------------------------------------------------------------
1. INTRODUCTION
Cloud computing [1] provides a “computing-as- a-service”
model in which compute resources are made available as a
utility service — an illusion of availability of as much
resources (e.g., CPU, memory, and I/O) as demanded by the
user. Moreover, cloud users pay only for the amount of
resources (a “pay-as-use” model) used by them. This model is
different from earlier infrastructure models, where enterprises
would invest huge amounts of money in building their own
computing infrastructure. Generally, traditional data centers
are set to meet the peak demand, which results in wastage of
resources during non- peak periods. To mitigate the above
problem, modern-day data centers are shifting to the cloud.
However, implementing cloud-based data centers requires a
great deal of flexibility and agility. For example, the dynamic
scaling and shrinking requirement needs compute resources to
be made available at very short notice. When computing
hardware is overloaded, it may be required to dynamically
transfer some of its load to another machine with minimal
interruption to the users. Virtualization technology can
provide these kinds of flexibilities.
We discuss the use of virtual machine migration [2] for
dynamic resource management in virtualized-based cloud
systems. As mentioned earlier migration is the process of
transferring state of a virtual machine (VM) from one physical
machine (PM) to another. Different techniques of migration
exits such as suspend-and-copy, pre- copy and post-copy. In
suspend-and-copy virtual machine is suspended and copies all
its pages and resumes at the destination machine. In pre-copy
method it transfers all its pages to the destination without
suspending the virtual machine. Once all the necessary pages
are transferred VM at the source is suspended and resumes at
the new source (destination). Live migration aim to minimize
the downtime of virtual machine either by transferring pages
before the machine gets suspended or copying minimal state
(post-copy) to start the VM and using demand-paging over the
network to fetch the remaining state.
In the current cloud computing environments, VM resource
scheduling only considers the current system condition and
ignores the previous state of system which causes the system
load imbalance. Number of VM migrations is more when most
of the load balancing takes place. The entire migration cost
becomes a problem when most of the VMs are migrated. So
it’s necessary to minimize the migration of VMs so that we
could radically improve the performance of the entire system
and saves a much amount of migration cost. So our proposed
prediction algorithm provides a solution to the mentioned
problem and also embraces multiple aspects and provides an
insight into their interactions of today’s cloud centers. The
main features of our algorithm can be listed as follows:
 Resource usage statics of each VM;
 Predicting job completion time;
 Cascading in migration in avoided;
 Minimizes overloading conditions;
The rest of the paper is organized as follows: In section 2, we
survey related work in dynamic resource allocation and live
migration in cloud computing environment. In section 3 we
will discuss about the proposed system. In section 4 we do the

__________________________________________________________________________________________
performance analysis on our proposed algorithm. Finally, in
section 4 summarize our findings and conclude the paper.
2. RELATED WORK
Cloud computing has attracted considerable research attention,
but only a small portion of the work has been done so far.
Many research works carried out in the field of resource
allocation and migration. Here, we survey those that proposed
certain methods and models for migration and resource
allocation.
In [3], the authors studied about resource allocation
mechanism that takes place in VM-based data centers. Here
they introduce two-tiered on demand resource allocation
mechanism that differs from traditional resource allocation
mechanisms in adding resource management level for VMs.
They proposed on demand resource allocation algorithms and
models to make resource allocation more dynamic. But they
didn’t mention any method to reduce the overloading
condition in virtual machines.
In [4], discusses about the framework that is used in managing
the clusters of virtual machines. Even though it implements
basic management mechanisms such as creating, destroying,
and migrating virtual machines, it doesn’t mention any method
how to minimize the migration of virtual machines.
In [5] and [6], author’s mentions about the priority based
resource allocation and threats that exits in virtual migration
respectively. [7] discusses the design options for migrating
OS’s running services with liveness constraints, focusing on
data center and cluster environments. Author’s discusses in
detail about the different phases in virtual machine migration.
In [8], author discusses about the dynamic resource allocation
in cloud environment by using VM migration. They present a
system that use virtualization technology to allocate data
center resources dynamically based on application demands
and support green computing by optimizing the number of
servers used. It explains about the situation in which VM get
migrated. A VM is migrated when there is hot spot detection.
A server is defined as a hot spot when its resource utilization
is above a hot threshold value. This indicates that server is
overloaded and some of its VMs running on it should be
migrated away.
As can be seen, no reported works covers the aspect of
minimizing the VM migration. This has motivated us to
develop a new algorithm to choose the destination node
(server) in such a way that there won’t be any cascade in
migration. In reference to [8], we also modified the algorithm
that to decide when to migrate and which VM to migrate in
the way mentioned below.
3. PROPOSED WORK
3.1 When to Migrate
There are many situations when migration of VMs becomes
necessary to maintain the overall efficiency of the data center.
These situations can be hot spot, periodic, load imbalance, and
addition of VMs. We are considering the situation of hot spot
over here. As we discussed earlier a server is said to be hot
spot if its resource utilization is above a threshold value. Let
that threshold value be HT. Now for every node, Ni, we need
to find the load of each node and let it be L(Ni). Let the total
sum of loads be SL. Now we will calculate the average of
loads present in all nodes and introduce another factor called
hot spot factor (α). Based on average value and hot spot factor
we determine the threshold value. So nodes with loads greater
than the threshold value will be considered as hot spot nodes.
Step by step algorithm is shown below.
Algorithm
 For each node Ni find load of Ni, i.e. Find L(Ni);
 Set SL=0, α be hot spot factor with value 1.2;
 For each node Ni do the following;
 SL=SL+L(Ni);
 Average (avg)=SL/Ni;
 HT=α * avg; end for;
 For each node Ni repeat up to step 9;
 If L(Ni) > HT then proceed to next step;
 Mark the node as hot spot node; end for;
It is not necessary that all hot spot nodes need to migrate. We
will perform the following algorithm on hot spot nodes to
determine which node to migrate.
3.2 Which to Migrate
Selecting one or more VMs for migration is a crucial decision
of the resource management heuristics. The migration process
not only makes the VMs unavailable for a period of time but
also consumes resources like network, CPU on source and
destination server. So it is important to make the correct
decision in choosing which VM to migrate.
Here we would consider only those nodes that are hot spot
nodes. Let HN be the list of hot spot nodes. A node or VM is
not considered for migration if it is already a migrated one.
Here in this algorithm we setup a three threshold values such
as threshold-input-length (TIL), min-threshold-input-length
(MTIL) and threshold time T. For every node or VM in HN
we consider both the total balance input and processed input
percentage. We select those nodes or VM with balance input
greater than the TIL value and processed input lengths less
than MTIL for migration. Step by step algorithm is shown
below.

__________________________________________________________________________________________
Algorithm
 Initially set values to TIL, MTIL and threshold time T;
 If current-last-migration time > T then go to next step;
 For every node in HN do the following;
 If node (i) not migrated go to next step;
 If balance input (i) > TIL go to next step;
 If processed input percentage (i) < MTIL go to next
step;
 Select the node (i) for migration; end for;
3.3 Where to Migrate
During migration destination PM should have enough
resources so that it can support incoming migrating VM. Here
we will discuss about the MVMM algorithm used to select the
destination for migrating VMs. The core part of MVMM
algorithm depends upon a VM allocation matrix (VA). So let’s
see VA matrix in detail.
In VA matrix each column represents VMs in the order of job
completion time. This completion time can be calculated on
the basis of speed at which each job on the VMs is processed.
Each row of VA matrix represents the number of PMs in the
entire system. Each value in matrix, say VA[i][j], represents
the number of loads pending in PM[i] when VM[j] terminates.
Now we will find the HT value using the same method
mentioned in the algorithm when to migrate. Now we find the
number of overloaded nodes on the basis of HT value for each
column in VA matrix. We repeat this step for every column
and calculate the sum of all overloaded nodes and are set as
Migrating Index (MI) value. This MI value plays a pivot role
in MVMM algorithm.
The above mentioned VA matrix can be explained with an
example. Let VM1, VM2, VM3 and VM4 represents virtual
machines in the column of the matrix and is in the order of job
completion time. Let PM1, PM2 and PM3 represents the
physical machines in the rows of VA matrix.
Table -1: VA matrix
Each value in matrix represents the number of loads pending
in each PMs. Here the HT value at the time of termination
VM1 is 3.6 (HT=avg * α). Here the avg = 3 ((2+4+3)/3) and
value of α is set as 1.2. So at the time VM1 terminates number
of nodes that exceeds the HT value is 1. This value is set in the
last row.
Now let’s see how algorithm works. Let P1, P2.., Pn be the
nodes (PM) present in the system. Let MV be the virtual
machine to migrate. For each node Pi, let’s assume that MV is
allocated to node Pi. Now we find the Migration Index (MI)
value for every allocation using VA matrix. After calculating
all the MI values, we find the node Pi with minimum MI
value. So we allocate MV to that node. Hence in following
this algorithm we could minimize the migration of VMs. Step
by step algorithm is shown below.
MVMM Algorithm
 Let P1, P2,…,Pn be the nodes(PM);
 MV is the VM to migrate;
 For each node Pi, assume MV is allocated to Pi;
 Calculate Migration Index (MI);
 End for;
 Allocate MV to node with minimum MI value;
4. PERFORMANCE ANALYSIS
The resulting algorithms have been implemented and a graph
has been plotted. From the table 1, it is clear that Migration
Index (MI) is a measure of future migration. And our
algorithm considers a different allocation sequence and finally
select sequence with minimum MI. But in the case of
algorithms used in previous works select any of sequence
without considering MI. So as the number of VMs increases
there is a drastic increase in migration using normal migration
algorithms that leads to a situation that we cannot manage it.
But in our algorithm, migration increases only linearly with
increase in VMs.
From the graph plotted it is proved that our MVMM algorithm
minimizes the virtual machine migration in cloud environment
and there by increases overall performance of the cloud
system.VM1 VM2 VM3 VM4
PM1 2 3 1 2
PM2 4 2 3 3
PM3 3 3 1 4
HT 3.6 3.2 2 3.6
No: of
overloaded
nodes
1 0 1 1 MI=3

__________________________________________________________________________________________
Fig -4: Performance comparison graph
5. CONCLUSIONS
In this position paper, we present a novel approach to
minimize virtual machine migration in cloud computing
environment. Our modified approach reduces migration
overhead up to 75% and the above graph plotted is a proof for
that. We only concentrate on minimizing VM migration and
eliminate starvation. But we are not considering time
complexity factor of our algorithm MVMM
ACKNOWLEDGEMENTS
I would like to express my gratitude to all those who gave me
help to complete this project. A special thanks to my guide
Prof. P Mohamed Shameem, H.O.D., CSE, TKM Institute of
Technology.
I am also thankful to staffs of the institution for guiding me
and providing me superior computing facilities. Last but not
least I would like to thank almighty for making this project a
reality.
REFERENCES
[1] M. Armbrust et al., “A view of Cloud Computing,”
Commun. ACM, vol. 53, no. 4, 2010, pp.50-58.
[2] Michael Nelson, “Fast Transparent Migration for
Virtual Machines.” 2005
[3] Ying Song et al., “A Two- Tiered On- Demand
Resource Allocation Mechanism for VM-Based Data
Centers”. 2013 IEEE
[4] M. McNett, D. Gupta, A. Vahdat, and G.M. Voelker,
“Usher: An Extensible Framework for Managing
Clusters of Virtual Machines,” Proc. Large Installation
System Administration Conf. (LISA ’07), Nov. 2007.
[5] Chandrasekhar S. Pawar and Rajnikant B. Wagh,
“Priority Based Dynamic resource allocation in Cloud
Computing,”
[6] Jon Oberheide, Evan Cooke, and Farnam Jahanian, “
Emperical Exploitation of Live Virtual Machine
Migration,”.
[7] Christopher Clark. 2005 “Live Migration of Virtual
Machines”. Proc. Symp. Networked Systems Design
& Implementation.
[8] Zhen Xiao, Weijia Song, and Qi Chen, “Dynamic
Resource Allocation Using Virtual Machines for Cloud
Computing Environment”. 2013 IEEE
[9] T. Wood, P. Shenoy, A. Venkataramani, and M.
Yousif, “Black-Box and Gray-Box Strategies for
Virtual Machine Migration,” Proc. Symp. Networked
Systems Design and Implementation (NSDI ’07), Apr.
2007.
[10] C.A. Waldspurger, “Memory Resource Management in
VMware ESX Server,” Proc. Symp. Operating Systems
Design and Implementation (OSDI ’02), Aug. 2002.
[11] G. Chen, H. Wenbo, J. Liu, S. Nath, L. Rigas, L. Xiao,
and F. Zhao, “Energy-Aware Server Provisioning and
Load Dispatching for Connection-Intensive Internet
Services,” Proc. USENIX Symp. Networked Systems
Design and Implementation (NSDI ’08), Apr. 2008.
[12] N. Bobroff, A. Kochut, and K. Beaty, “Dynamic
Placement of Virtual Machines for Managing SLA
Violations,” Proc. IFIP/IEEE Int’l Symp. Integrated
Network Management (IM ’07), 2007.
[13] J.S. Chase, D.C. Anderson, P.N. Thakar, A.M. Vahdat,
and R.P. Doyle, “Managing Energy and Server
Resources in Hosting Centers,” Proc. ACM Symp.
Operating System Principles (SOSP ’01), Oct. 2001.
[14] C. Tang, M. Steinder, M. Spreitzer, and G. Pacifici, “A
Scalable Application Placement Controller for
Enterprise Data Centers,” Proc. Int’l World Wide Web
Conf. (WWW ’07), May 2007.
[15] M. Zaharia, A. Konwinski, A.D. Joseph, R.H. Katz,
and I. Stoica, “Improving MapReduce Performance in
Heterogeneous Environments,” Proc. Symp. Operating
Systems Design and Implementation (OSDI ’08), 2008.
[16] M. Isard, V. Prabhakaran, J. Currey, U. Wieder, K.
Talwar, and A. Goldberg, “Quincy: Fair Scheduling for
Distributed Computing Clusters,” Proc. ACM Symp.
Operating System Principles (SOSP ’09), Oct. 2009.
[17] M. Zaharia, D. Borthakur, J. Sen Sarma, K. Elmeleegy,
S. Shenker, and I. Stoica, “Delay Scheduling: A Simple
Technique for Achieving Locality and Fairness in
Cluster Scheduling,” Proc. European Conf. Computer
Systems (EuroSys ’10), 2010.
[18] T. Sandholm and K. Lai, “Mapreduce Optimization
Using Regulated Dynamic Prioritization,” Proc. Int’l
Joint Conf. Measurement and Modeling of Computer
Systems (SIGMETRICS ’09), 2009.
[19] A. Singh, M. Korupolu, and D. Mohapatra, “Server-
Storage Virtualization: Integration and Load Balancing
in Data Centers,” Proc. ACM/IEEE Conf.
Supercomputing, 2008.
[20] Y. Toyoda, “A Simplified Algorithm for Obtaining
Approximate Solutions to Zero-One Programming
Problems, Management Science, vol. 21, pp. 1417-
1427, Aug. 1975.

__________________________________________________________________________________________
[21] R. Nathuji and K. Schwan, “Virtualpower: Coordinated
Power Management in Virtualized Enterprise
Systems,” Proc. ACM SIGOPS Symp. Operating
Systems Principles (SOSP ’07), 2007.
[22] D. Meisner, B.T. Gold, and T.F. Wenisch, “Powernap:
Eliminating Server Idle Power,” Proc. Int’l Conf.
Architectural Support for Programming Languages and
Operating Systems (ASPLOS ’09), 2009.
[23] Y. Agarwal, S. Hodges, R. Chandra, J. Scott, P. Bahl,
and R. Gupta, “Somniloquy: Augmenting Network
Interfaces to Reduce Pc Energy Usage,” Proc. USENIX
Symp. Networked Systems Design and Implementation
(NSDI ’09), 2009.
[24] T. Das, P. Padala, V.N. Padmanabhan, R. Ramjee, and
K.G. Shin, “Litegreen: Saving Energy in Networked
Desktops Using Virtualization,” Proc. USENIX Ann.
Technical Conf., 2010.
[25] Y. Agarwal, S. Savage, and R. Gupta, “Sleepserver: A
Software- Only Approach for Reducing the Energy
Consumption of PCS within Enterprise Environments,”
Proc. USENIX Ann. Technical Conf., 2010.
[26] N. Bila, E.d. Lara, K. Joshi, H.A. Lagar-Cavilla, M.
Hiltunen, and M. Satyanarayanan, “Jettison: Efficient
Idle Desktop Consolidation with Partial VM
Migration,” Proc. ACM European Conf. Computer
Systems (EuroSys ’12), 2012.

Enhancing minimal virtual machine migration in cloud environment

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Enhancing minimal virtual machine migration in cloud environment (20)

More from eSAT Publishing House (20)

Recently uploaded (20)

Enhancing minimal virtual machine migration in cloud environment