GRAPE: GPU-accelerated Zero-copy IoT Measurement Data Compression Transmission
Xintan Dou, Wendi Feng, J.-C. Liu · 2024
Nowadays, a large number of sensors are involved in the Internet of Things (IoT) systems, incurring significant amount of data that are required to be efficiently transmitted under the stringent latency and IoT network link capacity constraints. Compression is an efficient method to alleviate it. However, existing IoT gateways, especially those on the cloud, leverage the gateway Central Processing Unit (CPU) which incurs impaired performance due to data compression and transmission. To this end, we propose Grape, the Graphics Processing Unit (GPU)-based data compression solution for efficient IoT measurement data transmission. Under the hood, Grape consolidates multiple measurement data of the same kind as a video and compresses the video using hardware instructions provided by the GPU to significantly improve the data compression performance. Besides, Grape creates a direct data path between the Network Interface Card (NIC) and the host to avoid data copy between kernel space and user space. Our primary experimental results indicate that Grape is 5 × faster than the traditional approach.