الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 113 |  |  |
| | | from | 113 | | |
Abstract
Cloud computing has become the most important technology for commercial enterprises in recent years. As a result, the service provider’s objective is to keep costs low to provide a low-cost service in order to attract more customers. This aim can be achieved by utilizing the multi-tenant SaaS architecture, in which tenants share a common application instance and a shared database instance to store their data. Today’s tenants expect strict assurances for the performance and availability of rented services, known as performance service level agreements (SLAs). On the other side, the market forces providers to decrease operating expenses while also maximizing the usage of their hardware and software resources. In addition, the tenants’ workload patterns are irregular, requiring graceful adjustments due to unanticipated workload changes and variability.
As a result, in this thesis, we firstly present a Clustered-Based Multi-Tenant Database Management System (CB-MT DBMS) that makes use of the proposed sophisticated dynamic proactive provisioning technique for response time prediction. Furthermore, the proposed CB-MT DBMS uses the prediction results to anticipate the requirement for replication and migration in order to avoid SLA violations. Secondly, a new dynamic proactive provisioning technique has been proposed in order to deal with irregular workload issues that may lead to serious SLA violations and contractual penalties for that violation. The proposed technique uses forecasting models to forecast tenant query response time. Thirdly, a Multi-Tenant Migration Algorithm (MT-M) and an Enhanced Multi-Tenant Migration Algorithm (EMT-M) are presented which migrate the violated tenants on an elastic cluster of machines to mitigate the SLA violations. Fourthly, a Dynamic Proactive Provisioning Multi-Tenant Database Migration and Replication (MTDB-MR) algorithm is presented, which designed to reduce collisions and inconsistencies between replication and migration decisions for a group of violating tenants. Finally, a full dynamic proactive Multi-Tenant Database Allocation Algorithm (MT-A) is proposed in order to allocate the migrated and replicated violated tenant databases.
Experimental results prove that the proposed proactive provisioning technique created
using the Recurrent Neural Network (RNN) prediction model can reduce the value of the
prediction error more than its counterparts. Also, it shows that the proposed MT-M algorithm and the proposed EMT-M algorithm, results in a significant decrease in the total number of SLA violations and the number of client site violations over the previous algorithms. Furthermore, the experimental results showed that the proposed MTDB-MR algorithm is an ideal candidate for migration and replication of violating multi-tenant databases, as it can detect the source of the violation in advance and choose the optimum replication or migration decision for the application based on the results inferred from access log analysis services. In addition, it shows that the proposed MT-A algorithm chooses the optimal site to migrate or replicate the violating tenant according to a set of proposed rules, which leads to a significant reduction in both SLA violations and the time to execute the tenant’s transactions for more than one site and not just one site as in previous works.