The provisioning of on-demand cloud services has revolutionized the IT industry. This emerging paradigm has drastically increased the growth of data centers (DCs) worldwide. Consequently, this rising number of DCs is contributing to a large amount of world total power consumption. This has directed the attention of researchers and service providers to investigate a power-aware solution for the deployment and management of these systems and networks. However, these solutions could be beneficial only if derived from a precisely estimated power consumption at run-time. Accuracy in power estimation is a challenge in virtualized environments due to the lack of certainty of actual resources consumed by virtualized entities and of their impact on applications’ performance. The heterogeneous cloud, composed of multi-tenancy architecture, has also raised several management challenges for both service providers and their clients. Task scheduling and resource allocation in such a system are considered as an NP-hard problem. The inappropriate allocation of resources causes the under-utilization of servers, hence reducing throughput and energy efficiency. In this context, the cloud framework needs an effective management solution to maximize the use of available resources and capacity, and also to reduce the impact of their carbon footprint on the environment with reduced power consumption. This thesis addresses the issues of power measurement and resource utilization in virtualized environments as two primary objectives. At first, a survey on prior work of server power modeling and methods in virtualization architectures is carried out. This helps investigate the key challenges that elude the precision of power estimation when dealing with virtualized entities. A different systematic approach is then presented to improve the prediction accuracy in these networks, considering the resource abstraction at different architectural levels. Resource usage monitoring at the host and guest helps in identifying the difference in performance between the two. Using virtual Performance Monitoring Counters (vPMCs) at a guest level provides detailed information that helps in improving the prediction accuracy and can be further used for resource optimization, consolidation and load balancing. Later, the research also targets the critical issue of optimal resource utilization in cloud computing. This study seeks a generic, robust but simple approach to deal with resource allocation in cloud computing and networking. The inappropriate scheduling in the cloud causes under- and over- utilization of resources which in turn increases the power consumption and also degrades the system performance. This work first addresses some of the major challenges related to task scheduling in heterogeneous systems. After a critical analysis of existing approaches, this thesis presents a rather simple scheduling scheme based on the combination of heuristic solutions. Improved resource utilization with reduced processing time can be achieved using the proposed energy-efficient scheduling algorithm.

Power Modeling and Resource Optimization in Virtualized Environments

ABDUL SALAM, HUMAIRA
2020-02-26

Abstract

The provisioning of on-demand cloud services has revolutionized the IT industry. This emerging paradigm has drastically increased the growth of data centers (DCs) worldwide. Consequently, this rising number of DCs is contributing to a large amount of world total power consumption. This has directed the attention of researchers and service providers to investigate a power-aware solution for the deployment and management of these systems and networks. However, these solutions could be beneficial only if derived from a precisely estimated power consumption at run-time. Accuracy in power estimation is a challenge in virtualized environments due to the lack of certainty of actual resources consumed by virtualized entities and of their impact on applications’ performance. The heterogeneous cloud, composed of multi-tenancy architecture, has also raised several management challenges for both service providers and their clients. Task scheduling and resource allocation in such a system are considered as an NP-hard problem. The inappropriate allocation of resources causes the under-utilization of servers, hence reducing throughput and energy efficiency. In this context, the cloud framework needs an effective management solution to maximize the use of available resources and capacity, and also to reduce the impact of their carbon footprint on the environment with reduced power consumption. This thesis addresses the issues of power measurement and resource utilization in virtualized environments as two primary objectives. At first, a survey on prior work of server power modeling and methods in virtualization architectures is carried out. This helps investigate the key challenges that elude the precision of power estimation when dealing with virtualized entities. A different systematic approach is then presented to improve the prediction accuracy in these networks, considering the resource abstraction at different architectural levels. Resource usage monitoring at the host and guest helps in identifying the difference in performance between the two. Using virtual Performance Monitoring Counters (vPMCs) at a guest level provides detailed information that helps in improving the prediction accuracy and can be further used for resource optimization, consolidation and load balancing. Later, the research also targets the critical issue of optimal resource utilization in cloud computing. This study seeks a generic, robust but simple approach to deal with resource allocation in cloud computing and networking. The inappropriate scheduling in the cloud causes under- and over- utilization of resources which in turn increases the power consumption and also degrades the system performance. This work first addresses some of the major challenges related to task scheduling in heterogeneous systems. After a critical analysis of existing approaches, this thesis presents a rather simple scheduling scheme based on the combination of heuristic solutions. Improved resource utilization with reduced processing time can be achieved using the proposed energy-efficient scheduling algorithm.
Virtual machines, power profiling, power modeling, regression models, virtual environment, performance counters, machine learning, cloud network, task scheduling, heuristic algorithms, resource optimization
File in questo prodotto:
File Dimensione Formato  
phdunige_4308768.pdf

accesso aperto

Descrizione: Main article
Tipologia: Tesi di dottorato
Dimensione 3.22 MB
Formato Adobe PDF
3.22 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11567/998108
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact