Research On High-Throughput,Low-Latency Message Forwarding Technology For Edge Computing Gateways

As bridges for data exchange between terminal devices and edge servers,gateways play an irreplaceable role in the field of Io T and edge computing.However,as the number of terminal devices and the amount of data generated by those devices have increased dramatically in recent years,the limited data forwarding capacity of the gateway has become a performance bottleneck in the Io T-edge computing architecture.In order to effectively improve the data throughput of the link between terminal devices and edge servers,this thesis conducts research on high-throughput,low-latency message forwarding techniques for Io T edge gateways.The main works in this thesis are as follows:Design a user-space device driver for a class of PCI/PCIe network devices in edge computing gateways.The traditional kernel device driver results in low packet transmitting and receiving efficiency due to its complex packet delivery path in the kernel and frequent system calls.To address these issues,the user-space driver places the packet forwarding behaviour in the user space of the gateway operating system to achieve high throughput and low latency in the message forwarding process by bypassing the kernel and simplifying the packet delivery path within the gateway.Various performance optimization methods are designed for the above driver.For the small packet-intensive scenarios typical of edge computing environments,this thesis proposes a packet assembly method that further improves the data throughput of the gateway in this scenario by packetizing and forwarding packets with the same destination address and the same protocol type.For the abruptly changing data traffic in the edge network environment,this thesis introduces an adaptive packet assembly size algorithm for the packet assembly method,which is used to balance the data throughput of the gateway and the average forwarding delay of individual packets.Finally,for message forwarding between gateways,a deep reinforcement learning-based message routing strategy is proposed to help gateways choose a less loaded and lower delayed message forwarding path for next-hop delivery of packets.Performance evaluations show that user-space driver can achieve 27-34 times higher throughput and significantly lower average packet forwarding latency in the gateway compared to the traditional kernel network device driver,significantly improving the gateway’s message forwarding capabilities.Performance optimization methods such as the adaptive packet assembly method and the deep Q-network-based message routing strategy can also help the user-space driver to take advantage of their performance in specific scenarios.