Virtual Machine Placement for Supporting Network Cloud Services (bibtex)
by Eleni Kavvadia
Abstract:
A cloud computing network contains thousands of data centers or fog devices which can be found anywhere in the world and host millions of virtual machines instantiating cloud services to end users. These services can be hosted in one or more virtual machines that may be in a single or multiple data centers or fog devices. However, as the number of users and cloud services increase, key optimization problems become a priority for cloud service providers who focus on scalability problems in order to minimize the overall cost and eventually to lead to an increase in their profits. In this context, the problem of finding the number and location of virtual machines that instantiate a cloud service is a demanding problem, as its solution contains a large number of criteria and different possible formulations that will have to be studied. In a cloud network, firstly, a large number of virtual machines are randomly placed within the cloud computing network or in a manner which is adapted to the needs of the network traffic load. Therefore, the optimal position of virtual machines is an optimization problem since it requires global knowledge in a dynamic environment, such as that of a cloud network, as every time a change in the traffic load of the network appears, the collection of information and redefinition of the optimal values must be done in very short time. Obviously, such centralized approaches do not scale by the number of virtual machines and the cloud computing network size. Therefore, the need to create feasible and effective low-complexity solutions arises, even at the expense of an unguaranteed optimal solution. This dissertation presents approaches that move, replicate, or merge services to target reduction in total costs, based on network topologies and information solely available locally, ensuring the optimized use of services. First of all, the S-CORE (Scalable Communication Cost Reduction) policy is presented, a scalable virtual machine migration algorithm that dynamically reallocates virtual machines to servers, achieving both minimization of the overall communication cost and reduction of the communication of over-subscribed links in the core of a data center network. The simulations and implementation results show that S-CORE achieves a significant decrease in communication cost. In addition, the s-UFL (scalable-Uncapaciated Facility Location) policy is presented which uses virtual machine replication and merging, along with migration in order to optimize the location of the services in the network. The efficiency of this elastic policy and its characteristics are studied and analyzed thoroughly verifying that the reduction of the overall cost in the network is possible under certain conditions. The ease of implementation in the network nodes (data centers and fog devices) is another advantage of this policy. Note, that the facility location problem in this case is studied as an uncapacitated case. Nevertheless, the results of this research can easily be extended to apply to capacitated scenarios.
Reference:
Eleni Kavvadia, "Virtual Machine Placement for Supporting Network Cloud Services", Ph.D. Thesis, Ionian University, 2017.
Bibtex Entry:
@phdthesis{kavvadia2017thesis,
	Abstract = {A cloud computing network contains thousands of data centers or fog devices which can be found anywhere in the world and host millions of virtual machines instantiating cloud services to end users. These services can be hosted in one or more virtual machines that may be in a single or multiple data  centers or fog devices. However, as the number of users and cloud services increase, key optimization problems become a priority for cloud service providers who focus on scalability problems in order to minimize the overall cost and eventually to lead to an increase in their profits. In this context, the problem of finding the number and location of virtual machines that instantiate a cloud service is a demanding problem, as its solution contains a large number of criteria and different possible formulations that will have to be studied. In a cloud network, firstly, a large number of virtual machines are randomly placed within the cloud computing network or in a manner which is adapted to the needs of the network traffic load. Therefore, the optimal position of virtual machines is an optimization problem since it requires global knowledge in a dynamic environment, such as that of a cloud network, as every time a change in the traffic load of the network appears, the collection of information and redefinition of the optimal values must be done in very short time. Obviously, such centralized approaches do not scale by the number of virtual machines and the cloud computing network size. Therefore, the need to create feasible and effective low-complexity solutions arises, even at the expense of an unguaranteed optimal solution. This dissertation presents approaches that move, replicate, or merge services to target reduction in total costs, based on network topologies and information solely available locally, ensuring the optimized use of services. 
	First of all, the S-CORE (Scalable Communication Cost  Reduction) policy is presented, a scalable virtual machine migration algorithm that dynamically reallocates virtual machines to servers, achieving both minimization of the overall communication cost and reduction of the communication of over-subscribed links in the core of a data center network. The simulations and implementation results show that S-CORE achieves a significant decrease in communication cost.
	In addition, the s-UFL (scalable-Uncapaciated Facility Location) policy is presented which uses virtual machine replication and merging, along with migration in order to optimize the location of the services in the network. The efficiency of this elastic policy and its characteristics are studied and analyzed  thoroughly verifying that the reduction of the overall cost in the network is possible under certain conditions. The ease of implementation in the network nodes (data centers and fog devices) is another advantage of this policy. Note, that the facility location problem in this case is studied as an uncapacitated case. Nevertheless, the results of this research can easily be extended to apply to capacitated scenarios.},
	Author = {Kavvadia, Eleni},
	Date-Modified = {2020-01-27 23:15:59 +0200},
	Keywords = {stphdthesis},
	Month = {June},
	Note = {Text in greek},
	School = {Ionian University},
	Title = {{{Virtual Machine Placement for Supporting Network Cloud Services}}},
	Type = {{Ph.D. Thesis}},
	Year = {2017}}
Powered by bibtexbrowser