Multi-GPU Acceleration for Smart Grid Data Compression

Smart grids that utilize digital computation and communication to monitor and control the power distribution and usage have become one of important technologies to guard the efficiency and reliability of electricity services. The combination of cloud systems makes smart grids even more convenient for data acquisition, processing, management, and accessing. However, the massive data, automatically generated by meters, can easily exhaust the computational resources and therefore slowdown the performance. Normal data compression methods, although achieving good compression ratio, do not allow the compressed data being queried and operated. The live data compression technique, by which the compressed data can still be queried, can resolve this problem. Nevertheless, such technique requires heavy computation and could interfere with regular data operations. In this paper, we propose an acceleration method for the live data compression for smart grid data using GPU (Graphics Processing Unit), which utilizes the ideas from machine learning and high performance computing. The accelerated program can achieve near 200 times speedups using four GPUs, by which hundred millions of smart meter records can be compressed in seconds.