Lossless Network Traffic Switching via Queue Congestion Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network devices face inefficiencies in data switching, leading to data loss and retransmission due to congested queues, which can degrade channel efficiency and user experience, and mechanisms like PAUSE and PFC often result in inefficient buffer utilization and dropped packets.
Innovation Solution
A network device system with a dedicated stream processor for AV traffic and a host processor for non-AV traffic, utilizing separate queues and congestion management to prevent data release to congested queues, ensuring lossless switching by prioritizing AV traffic and managing congestion status to minimize data loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If PAUSE or PFC mechanisms are deployed to prevent data loss, then data loss is reduced, but buffer utilization becomes inefficient and requires reserving at least two times the MTU per port/queue
Solution Approach 1:
The patent segments the buffer management by introducing separate admission control handlers for different traffic types (AV vs. non-AV traffic). Each handler independently monitors queue congestion and controls data release, allowing differentiated buffer management strategies without requiring excessive reserved buffer space across all queues.
Solution Approach 2:
The patent applies local quality by implementing congestion-dependent buffer allocation where the buffer management behavior varies based on local queue conditions. The admission control handler dynamically adjusts data release based on real-time congestion status of specific queues, rather than applying uniform buffer management across all traffic.
2Reliability
If PAUSE mechanism is deployed to control congestion, then data loss is reduced, but all traffic from the sourcing device is put in backpressure even when only a single egress queue is congested
Solution Approach 1:
The patent segments traffic control by implementing separate admission control handlers that independently manage different traffic types. When congestion occurs in one queue, only the affected traffic class is controlled, while other traffic classes continue flowing, avoiding the blanket backpressure effect of traditional PAUSE mechanisms.
Solution Approach 2:
The patent applies partial action by implementing fine-grained congestion control that acts only on the specific queue and traffic class experiencing congestion, rather than applying backpressure to all traffic. The admission control handler selectively prevents data release only for congested queues while allowing other queues to operate normally.
3Reliability
If PFC mechanism is deployed for per traffic class backpressure, then data loss is reduced, but the source device cannot send traffic to any other switch port queue of the same traffic class when a single queue is in backpressure
Solution Approach 1:
The patent segments the backpressure control by implementing independent admission control handlers for each queue rather than applying class-wide backpressure. This allows selective control where only the specific congested queue is blocked, while other queues in the same traffic class continue to receive and forward traffic normally.
4Reliability
If buffer space is reserved for lossless behavior (two times MTU per port/queue), then data loss is prevented, but buffer space cannot be used by queues with heavy traffic leading to dropped packets
Solution Approach 1:
The patent implements dynamic buffer management where the admission control handler continuously monitors queue congestion status and dynamically adjusts data release decisions. Buffer space is not statically reserved but dynamically allocated based on real-time traffic conditions, allowing queues with heavy traffic to utilize available buffer space when needed.
Solution Approach 2:
The patent enables queues to self-regulate their buffer usage through the admission control mechanism. Each queue's admission control handler independently monitors its own congestion status and controls data release, allowing efficient self-service buffer management without requiring pre-reserved buffer overhead.
Data Source
AI summary
A system for lossless switching of traffic in a network device may be implemented when a network switch is integrated into a gateway device, or with any other data source. A processor of the gateway device may receive queue depth information for queues of the network switch. The processor may prevent data from being transmitted to congested queues of the network switch, while allowing data to be transmitted to uncongested queues. In this manner, data loss can be avoided through the network switch for data sourced from the gateway device, such as audio-video data retrieved from a hard drive, audio-video data received from a tuner, etc. Furthermore, re-transmission at higher layers can be reduced. Since the subject system observes congestion for each individual queue, only traffic destined to that particular, congested, queue is affected, e.g. paused. Traffic to non-congested queues is not affected, regardless of traffic class or egress port.