Effects on performance and energy reduction by file relocation based on file-access correlations
Proceedings of the 2012 Joint EDBT/ICDT Workshops
| Hi-index | 0.00 |
The amount of electrical power consumed in data centers is increasing, so reducing power consumption is an important issue for cloud computing. An effective approach to reducing power consumption is suitable data allocation in data storage. Previous studies of this approach were mainly based on data-access frequency to decide the allocation of data. We propose a data allocation method based on query-pattern analysis. Because many Internet applications use the XML format to transfer and store data, we study XML queries here. Fortunately, a tree structure in XML is useful to distinguish the chunk of information to be accessed by a query. Our proposed algorithm, XARrP, an XML data Allocation algorithm for Reducing Power consumption, treats queries in XPath format with special symbols and attributes. XARP is separated into four steps: XPath processing, class mining, association, and data allocation. The XPath process retrieves multiple XML classes from query logs, even though some of them may not be expressed exactly. The class mining process analyzes query patterns to discover frequent classes. Infrequent classes are separated into several class sets in the association process. Finally, the frequent classes and the infrequent class sets are stored in different disks to reduce power consumption. Assuming no special hardware but disk drives spinning down after a specific time period, evaluation results show that power consumption is reduced by 32.5% compared with applying a naive striping method and reduced by 10.8% when compared with a method applying a previous XML cache algorithm. The performance expressed as number of transactions processed per unit of power of the proposed method is 2.5 times better than that of striping.