Mirroring Dropped Packets via Early Pipeline Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Network devices often drop packets due to various reasons such as queue congestion, MTU violations, or errors, making it challenging to understand the reasons behind packet drops, especially when packets have already been assigned to an egress port or data read from memory, as mirroring late in the processing pipeline is impractical.
Innovation Solution
The network device encapsulates to-be-dropped packets with an encapsulation header containing a destination address of a packet analyzer and feeds them back into the processing pipeline prior to egress port assignment, allowing the packets to be forwarded to the analyzer for analysis, along with relevant metadata for further investigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If packets are mirrored late in the processing pipeline (after egress port assignment or memory read), then it is possible to capture packets that would otherwise be dropped, but the complexity of implementing the mirroring mechanism increases significantly
Solution Approach 1:
The patent applies preliminary action by injecting a mirror copy of the packet back into the processing pipeline at an earlier stage (before egress port assignment) after the mirroring decision has been made. This allows the packet to be processed through the normal pipeline again and eventually delivered to the mirror port, avoiding the need for complex post-processing mirroring mechanisms. The mirror copy is created and re-injected before the packet would normally be dropped, ensuring it reaches the analysis destination through existing infrastructure.
2Productivity
If packets are dropped due to queue congestion or MTU violations, then network performance is maintained, but the ability to analyze why packets were dropped is lost
Solution Approach 1:
The patent applies copying by creating a mirror copy of packets that are destined to be dropped due to congestion, MTU violations, or other reasons. Instead of dropping the original packet and losing information about why it was dropped, a copy is made and redirected to a mirror port where analysis can occur. The original packet is still dropped to maintain network performance, while the copy preserves the information needed for analysis of drop reasons.
Solution Approach 2:
The patent implements feedback by delivering mirror copies of dropped packets to a designated mirror port where network operators can analyze them. This creates a feedback loop where information about packet drops (including reasons such as congestion, MTU violations, or policy decisions) is returned to operators, enabling them to understand and address the root causes of packet drops while the original dropping behavior continues to maintain network performance.
3Measurement precision
If packets are mirrored after egress port assignment, then accurate drop analysis is possible, but the mirroring operation becomes more complex and resource-intensive
Solution Approach 1:
The patent applies preliminary action by performing the mirroring operation at a point in the processing pipeline where the decision to drop the packet has already been made, but before the packet is actually discarded. By re-injecting the mirror copy back into the pipeline at this stage, the system leverages the already-completed processing decisions while avoiding the need for complex post-processing operations. This timing optimizes resource usage by utilizing existing processing state rather than requiring additional intensive operations.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In one embodiment, a network device (10), includes ports (12) to serve as ingress ports and egress ports, a memory to store received network packets, a switch fabric (14), a controller to monitor queues of the egress ports and make a decision to drop at least a network packet of the received network packets, the network packet having a first destination address associated with a first network node, and mirroring circuitry (18) to encapsulate the network packet with an encapsulation header including a second destination address different from the first destination address, and feedback the encapsulated network packet into a processing pipeline of the received network packets within the network device at a point in the processing pipeline that is prior to egress port assignment, wherein the switch fabric (14) is configured to forward the encapsulated network packet to the packet analyzer responsively to the second destination address in the encapsulation header.