Layer 3 Fair Rate Congestion Control for Data Center Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional TCP/IP stacks in data centers fail to simultaneously provide high throughput and low latency, and existing congestion control mechanisms like PFC and QCN have limitations, particularly in IP-routed environments, leading to performance issues such as head-of-line blocking and unfairness.
Innovation Solution
The implementation of Layer 3 Fair Rate Congestion Notification (L3FCN) in network devices, which involves receiving data packets, updating fair rates using proportional integral control and multiplicative decrease, identifying elephant flows, and sending fair rate messages to their sources to manage queue lengths and prevent congestion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Priority-based Flow Control (PFC) is used to enable drop-free network, then network reliability is improved, but application performance deteriorates due to head-of-line blocking and unfairness
Solution Approach 1:
The patent implements explicit feedback mechanisms where network devices send congestion notifications back to sources. The feedback includes information about queue lengths and congestion states, enabling sources to adjust their transmission rates dynamically. This resolves the contradiction by providing controlled feedback that prevents head-of-line blocking while maintaining reliability through proactive congestion management.
Solution Approach 2:
The patent introduces dynamic rate adjustment mechanisms that allow transmission rates to change based on real-time network conditions. Sources can dynamically modify their sending rates in response to congestion notifications, transforming the static PFC approach into a dynamic system that adapts to changing traffic patterns, thereby improving application performance while maintaining network reliability.
2Reliability
If Quantized Congestion Notification (QCN) is implemented to solve PFC limitations, then congestion control effectiveness is improved, but applicability deteriorates in IP-routed data centers
Solution Approach 1:
The patent designs a congestion control mechanism that operates effectively across different network architectures including both Ethernet switched networks and IP-routed data centers. By implementing Layer 3 integration and using IP header information for flow identification, the system achieves universal applicability while maintaining the congestion control effectiveness of QCN-like approaches in diverse network environments.
Solution Approach 2:
The patent transitions from pure Layer 2 operation to Layer 3 integration by incorporating IP routing information and using TCP sequence numbers for flow identification. This dimensional shift from link-layer to network-layer operation enables the congestion control mechanism to function effectively in IP-routed data centers while preserving the core congestion notification capabilities.
3Adaptability or versatility
If traditional TCP/IP stacks are used for communication, then compatibility is improved, but simultaneous high throughput and low latency cannot be achieved
Solution Approach 1:
The patent introduces an intermediary congestion control layer that operates between the traditional TCP/IP stack and the network infrastructure. This intermediary mechanism provides explicit congestion notifications and rate control without replacing the existing TCP/IP protocol stack, thereby maintaining compatibility while achieving improved throughput and latency through proactive congestion management and rate limiting.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A network device, including ports that receive/send data packets from/to a network, receives data packets of multiple traffic flows, and populates a queue in memory with the data packets. The network device periodically updates a fair rate for the multiple traffic flows to converge a length of the queue to a reference length. Specifically, the network device determines a length of the queue, a change in the length from a previous length, and a deviation of the length from the reference length. The network device detects an increase in the change in length above a threshold that is based on the reference length. If the increase is not above the threshold, the network device derives the fair rate from a previous fair rate using proportional integral control. The network device identifies elephant flows among the multiple traffic flows, and sends the fair rate to a source of each elephant flow.