Packet Redirector Circuitry for Data Center Incast Congestion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In data center networks, incast occurrences lead to traffic overload at egress ports due to multiple spine switches sending traffic to the same downstream leaf switch port, causing queue congestion, increased latency, and packet drops, which existing solutions like Reorder Resilient Transport (RRT) address partially but still result in reduced throughput and higher latencies.
Innovation Solution
Implementing a packet burst-tolerant disaggregated incast absorption system that redirects packets from congested ports to non-congested ones on a leaf node, using network interface devices and servers for buffering, and applying Reorder Resilient Transport (RRT) at the destination to reorder out-of-order packets, thereby maintaining near-lossless Ethernet fabric performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple spine switches send traffic to the same downstream leaf switch port, then traffic throughput increases, but queue congestion and latency increase
Solution Approach 1:
The patent segments the queue buffer into multiple independent queue buffers, each associated with a specific egress port. This allows independent management and redirection of packets for different ports, enabling fine-grained congestion control where only affected port queues are redirected while others continue normal operation, thus maintaining throughput while reducing latency.
Solution Approach 2:
The patent introduces an intermediary redirection mechanism at the leaf switch that intercepts packets destined for congested ports and redirects them to alternative egress ports. This intermediary layer absorbs the congestion impact without affecting the entire system, allowing traffic to flow through alternative paths and reducing overall latency while maintaining throughput.
2Reliability
If congestion signaling is sent to slow down packet transmission, then incast occurrence is reduced, but packet throughput decreases
Solution Approach 1:
The patent performs preliminary congestion detection and packet redirection before queues become fully congested. By proactively identifying potential incast scenarios and redirecting packets in advance, the system prevents queue overflow and packet loss while maintaining steady throughput, avoiding the need to slow down transmission after congestion occurs.
Solution Approach 2:
The patent implements a feedback mechanism that monitors queue fullness levels and dynamically adjusts packet redirection decisions. When queues approach capacity, the system activates redirection; when queues are clear, normal transmission resumes. This feedback-based approach maintains reliability by preventing incast while preserving throughput through adaptive control.
3Reliability
If packets are redirected to alternative ports, then congestion is reduced, but packet delivery order is disrupted
Solution Approach 1:
The patent uses packet tagging (analogous to color changes) to mark redirected packets with identification information indicating their original destination and redirection status. This tagging allows the system to track and manage redirected packets separately, enabling proper reordering at the destination while maintaining congestion relief benefits.
Solution Approach 2:
The patent changes the routing parameter for redirected packets by modifying their egress port assignment while maintaining other packet characteristics. This parameter change enables flexible redirection to alternative ports while the system tracks these changes to restore proper packet ordering at the destination, balancing congestion management with delivery stability.
4Reliability
If queue capacity is increased to handle incast, then packet drops are reduced, but latency increases
Solution Approach 1:
The patent segments the buffer capacity into multiple port-specific queue buffers rather than using a single large buffer. This segmentation allows the system to prevent packet drops at individual congested ports while other ports maintain low-latency operation, achieving reliability without system-wide latency increase.
Solution Approach 2:
The patent extracts congested packets from the main queue flow and redirects them to alternative paths. By removing problematic packets from the congested queue before they can cause overflow and packet loss, the system maintains reliability without needing to increase overall queue capacity, thus avoiding latency penalties.
Data Source
AI summary
Examples described herein relate to a switch comprising: circuitry to detect congestion at a target port and re-direct one or more packets directed to the target port to one or more other ports for re-circulation via one or more uncongested ports based on congestion at the target port. In some examples, the circuitry is to identify the target port in the re-directed one or more packets. In some examples, the circuitry is to transmit a congestion level indicator to the one or more other ports based on a congestion level of the target port.