Lossless Ethernet Routing via Traffic Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Ethernet networks lack reliable link-layer transmission mechanisms for data packets, especially when integrated into IP networks, leading to packet loss during congestion, and require additional transport-layer protocols for reliable data transfer, complicating the implementation of protocols like RDMA over multiple subnets.
Innovation Solution
Implementing a protocol infrastructure that defines two classes of link-layer traffic, where data traffic is transmitted in a lossless class using Ethernet Priority Flow Control (PFC) and control traffic is transmitted in a lossy class, avoiding deadlocks by intentionally dropping packets during congestion, allowing for reliable data transfer across multiple subnets without relying on transport-layer protocols.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If Ethernet networks transmit data packets at high speed, then throughput is improved, but packet loss occurs when receivers run out of buffer space
Solution Approach 1:
The invention segments traffic into control packets and data packets, applying different handling mechanisms to each. Control packets use traditional best-effort delivery, while data packets benefit from lossless forwarding mechanisms, thereby resolving the contradiction between high throughput and reliable delivery.
Solution Approach 2:
The invention introduces an intermediary mechanism (lossless Ethernet service class) between the transmitter and receiver that manages buffer space allocation and packet forwarding. This intermediary ensures that data packets are not lost due to buffer overflow while maintaining high throughput.
2Reliability
If priority flow control is implemented for data traffic, then packet loss is prevented, but network complexity increases
Solution Approach 1:
The invention segments the network service into two distinct classes: best-effort service for control packets and lossless service for data packets. This segmentation allows PFC to be applied selectively only where needed, reducing overall network complexity while maintaining reliability for critical data traffic.
Solution Approach 2:
The invention applies different service qualities locally to different packet types. Control packets receive best-effort service with simple forwarding, while data packets receive enhanced lossless service with PFC mechanisms. This local differentiation maintains reliability where needed without adding complexity throughout the entire network.
3Speed
If control traffic is transmitted with high priority, then routing configuration speed is improved, but deadlocks may occur in cyclic network paths
Solution Approach 1:
The invention segments control traffic into a separate best-effort service class, distinct from data traffic. This segmentation allows control packets to be transmitted efficiently without the constraints of lossless forwarding, preventing deadlocks in cyclic paths while maintaining adequate configuration speed.
Solution Approach 2:
The invention inverts the conventional approach by not applying high priority and lossless guarantees to control traffic. Instead, control packets use best-effort delivery, which paradoxically improves network stability by avoiding the deadlocks that can occur with strict priority and buffer guarantees, while still achieving adequate configuration speed.
4Reliability
If lossless forwarding is applied to all traffic classes, then packet delivery reliability is improved, but buffer management complexity increases
Solution Approach 1:
The invention segments traffic into two classes and applies buffer management mechanisms selectively. Only data packets in the lossless service class receive buffer management and PFC treatment, while control packets in the best-effort class use simple buffer handling. This segmentation reduces overall buffer management complexity while maintaining reliability for data traffic.
Data Source
AI summary
A method for communication in a packet data network including at least first and second subnets interconnected by routers. The method includes defining at least first and second classes of link-layer traffic within the subnets, such that the link-layer traffic in the first class is transmitted among nodes in the network without loss of packets, while at least some of the packets in the second class are dropped in case of network congestion. The routers are configured by transmitting control traffic over the network in the packets of the second class. Data traffic is transmitted between the nodes in the first and second subnets via the configured routers in the packets of the first class.