| Data center (DC) operators face a bewildering set of choices when considering how to provision resources for data centers in ways that are cost-effective. DC operators have to figure out the impact of alternative hardware (including I/O, memory, CPU and networking), alternative OSs and alternative Database management system (DBMS) configurations. For example, modern I/O subsystems often have a rich mix of fast, high performing, but expensive SSDs sitting alongside with cheaper but relatively slower (for random accesses) traditional hard disk drives. On the OS and DBMS side, there is an ongoing debate about the performance issues when using virtual machines and databases, especially in multi-tenancy cloud hosted environments. On the I/O side, there are a number of open issues related to how DBMSs can best use modern (SSD-based) I/O systems in cost-effective ways. Data center operators need to take into account all these factors when provisioning resources for DBMSs and other cloud services, with an overall goal of minimizing the total operating cost (TOC) or TOC/performance of running the workloads. And, all this has to be done while abiding by potential Service Level Agreements (SLAs). The focus of this thesis is on looking at hardware provisioning and VM-based DBMS hosting solutions while minimizing TOC and meeting performance SLAs. In addition, this thesis also examines how to provision storage clouds (for applications such as database archiving) in cost-effective ways, while meeting performance SLAs. |
