Thesis etd-09222014-153512 |
Link copiato negli appunti
Thesis type
Tesi di laurea magistrale
Author
BUCCARELLA, MATTIA
email address
m.buccarella@gmail.com
URN
etd-09222014-153512
Thesis title
A MILP-based Approach to Multiple Cloud Resource Allocation
Department
INFORMATICA
Course of study
INFORMATICA E NETWORKING
Supervisors
relatore Prof. Coppola, Massimo
Keywords
- allocation
- broker
- milp
- multi-cloud
- optimization
Graduation session start date
10/10/2014
Availability
Withheld
Release date
10/10/2084
Summary
The thesis studies a model of resource allocation for Cloud applications that allows using multiple clouds at the same time for a large set of applications.
The model allows resource allocation to be cast as a Mixed/Integer Linear Programming problem, where optimal solutions can be computed subject to various types of platform constraints (e.g. application integrity when mapped across Clouds, resource capacity, application budget) and according to platform parameters (resource and migration costs). The objective functions takes then into account application cost as well as penalties for rejecting/migrating applications.
The model is encoded in the MathProg language, and a software prototype has been built. It exploits standard MILP optimizers GLPK and CBC and a set of ad-hoc tools in Java that allow to focus the optimization on a subset of the available resources and applications. The model and the prototype are evaluated with respect to their effectiveness and running time using a real life dataset made available by Google, tracing the execution of a very large number of Cloud jobs for more than 6 hours.
The model allows resource allocation to be cast as a Mixed/Integer Linear Programming problem, where optimal solutions can be computed subject to various types of platform constraints (e.g. application integrity when mapped across Clouds, resource capacity, application budget) and according to platform parameters (resource and migration costs). The objective functions takes then into account application cost as well as penalties for rejecting/migrating applications.
The model is encoded in the MathProg language, and a software prototype has been built. It exploits standard MILP optimizers GLPK and CBC and a set of ad-hoc tools in Java that allow to focus the optimization on a subset of the available resources and applications. The model and the prototype are evaluated with respect to their effectiveness and running time using a real life dataset made available by Google, tracing the execution of a very large number of Cloud jobs for more than 6 hours.
File
| Nome file | Dimensione |
|---|---|
Thesis not available for consultation. |
|