UDP Congestion Control via Overlay ECN Headers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current network congestion control methods are ineffective for UDP protocols in data center networks, leading to packet loss and inefficient bandwidth usage due to the lack of explicit congestion notification support, which affects time-sensitive data flows and bandwidth allocation.
Innovation Solution
The method involves encapsulating an outer overlay network packet header with an ECN identifier in the IP header, allowing for congestion control by forwarding UDP packets to a preset slow channel and adjusting bandwidth allocation dynamically based on network conditions, thereby reducing packet loss and retransmissions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If ECN technology is used for congestion control, then congestion control capability is improved, but it is only applicable to TCP protocol and does not support UDP protocol
Solution Approach 1:
The patent extends ECN congestion control capability from TCP protocol to UDP protocol by introducing a new header field structure that can carry congestion notification information for both protocols. The outer IP header includes an ECN field that can be set by network devices to notify both TCP and UDP traffic of congestion conditions, making the congestion control mechanism universal across different transport protocols.
Solution Approach 2:
The patent introduces an intermediate overlay network header structure that mediates between the underlying TCP/UDP protocols and the ECN congestion control mechanism. This intermediate header acts as a bridge, allowing ECN congestion notifications to be transmitted through the overlay network to both TCP and UDP packets, enabling UDP to benefit from ECN-based congestion control without modifying the original UDP protocol.
2Reliability
If packet loss is used for congestion detection in TCP, then congestion can be detected, but it causes long timeout periods and retransmission delays affecting time-sensitive data flows
Solution Approach 1:
The patent implements preliminary congestion notification by setting the ECN field in the outer IP header before packets experience loss. Network devices along the transmission path can proactively mark packets with congestion information in the ECN field, allowing the sender to detect congestion before packet loss occurs. This preliminary action eliminates the need for timeout-based detection and retransmission delays.
Solution Approach 2:
The patent establishes a feedback mechanism where network devices monitor network conditions and set the ECN field in packet headers to provide real-time congestion information back to the sender. This continuous feedback loop allows the sender to dynamically adjust transmission rates based on current network conditions without waiting for packet loss or timeout events, significantly reducing delays for time-sensitive applications.
Data Source
Figure 1~2
Figure 3~4
Figure 5~8
AI summary
This application provides a method for controlling network congestion, an access device, and a computer readable storage medium, where the method includes: overlaying an overlay network packet header on an encapsulation outer layer of a transmit packet, where the overlay network packet header includes an outer internet protocol IP header, and an ECN identifier of an explicit congestion notification technology is set in the outer IP header; decapsulating the overlay network packet header for an encapsulated reply packet, where an inner congestion identifier that is based on the ECN identifier is obtained from an IP header of the decapsulated reply packet through matching; and if the decapsulated reply packet is a user datagram protocol UDP packet, forwarding the UDP packet to a preset slow channel. After the embodiments of the present invention are used, the embodiments of the present invention are applicable to a UDP protocol, so that network congestion can be controlled.