Tuning servers, storage and database for energy efficient data warehouses

Undoubtedly, reducing power consumption is at the top of the priority list for system vendors, data center managers who are challenged by customers, analysts, and government agencies to implement green initiatives. Hardware and software vendors have developed an array of power preserving techniques. On-demand-driven clock speeds for processors, energy efficient power supplies, and operating-system-controlled dynamic power modes are just a few hardware examples. Software vendors have contributed to energy efficiency by implementing power efficient coding methods, such as advanced compression and enabling applications to take advantage of large memory caches. However, adoption of these power-preserving technologies in data centers is not straightforward, especially, for large, complex applications such as data warehouses. Data warehouse workloads typically have oscillating resource utilizations, which makes identifying the largest power consumers difficult. Most importantly, while preserving power remains a critical consideration, performance and availability goals must still be met with systems using power-preserving technologies. This paper evaluates the tradeoffs between existing power-saving techniques and their performance impact on data warehouse applications. Our analysis will guide system developers and data center managers in making informed decisions regarding adopting power-preserving techniques.