Logo image
A Live Data Compression Method for Sensor Data and Its GPU Implementation
Thesis

A Live Data Compression Method for Sensor Data and Its GPU Implementation

Chen, Zhi-Hung
Masters, 國立清華大學, 資訊工程學系
2012

Abstract

資料壓縮 即時查詢 資料庫 data compression GPU database
In-Snergy, a type of smart grid system, collects the electricity usage of the users’ appliances. The collected sensor data stored in the database increases day by day. These data becomes a huge load of the server storage. We propose a compression method for the sensor data. First, we divide the massive sensor data into many segments. Then we compress these segments and store the compressed segments in the database. The compressed data in the database can be queried by specific SQL. To compress the segments of the sensor data, we propose three kinds of compression methods—the dictionary method, the shift method and the incremental method. Based on the property of the segments, we choose the appropriate method to compress the segments. In our implementation, the size of compressed table is 29.44% of the size of the original table. The size of the database index of the compressed table is 3.87% of the index size of the original table. Because the process of compression is time-consuming, we use CUDA to accelerate the compression process. The speedup of each part of the process is 22 to 191 times in a 4-GPU environment.

Metrics

1 Record Views

Details

Logo image