Abstract

Cloud computing is a fast and growing paradigm for hosting applications services that belong to the Application Service Providers (ASPs). However, Quality of Service (QoS) remains an issue that opens different research areas in a distributed, elastic and dynamic
cloud platform. One major issue raised for the ASP is service High Availability (HA). Service availability is a non-functional requirement that indicates the period of time the service is provided for the end customer. Managing the availability of different application services during the runtime in a cloud cluster is not an easy task to do due to several challenges. The key success to maintain service availability is to provide a mechanism to protect service against failure and recover the service once the failure happens as fast as possible.
This thesis proposes a general framework for availability management and enables continuity for applications services in the cloud computing environment. We address service availability and propose efficient solutions from different perspectives. First, the thesis proposes a reactive framework that can maintain HA of the application service in a virtualized computing cluster. Second, a proactive service availability framework is proposed. The framework uses deep learning methods to predict application task termination status (Success or Fail) in cloud cluster using three public available datasets. The results show the used methods can predict task termination status with high accuracy. Third, a failure-aware task scheduler approach is proposed. The scheduler uses a heuristic approach to solve task scheduling NP-hard problem with the objective to minimize failure probability and resources usage of tasks. The results show the ability of the scheduler to protect many
tasks and save a large number of resources. Fourth, the thesis proposes an availability-aware Virtual Machine (VM) dynamic placement framework. The framework tackles VMs placement as a response to different request types that include deploying a new application, VM scaling, and migration. Moreover, an optimization approach that is based on the heuristic AntColony algorithm is proposed to solve the VM placement NP-hard problem. The approach targets multiple objectives to minimize power consumption, resources wastage, and failure of the active servers that are used to host the VMs. In addition, the placement approach tries to provide application service availability as close as possible to the requirements by ASP and avoids violation of the Service Level Agreement (SLA). The results show the ability of the framework to increase the admissibility of new applications that meet the availability requirements and enhance the resources utilization of servers, compared to the existing VM placement solutions in the literature.