Unified Final Buffer for Network Switch Traffic Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Switches face challenges in optimizing network traffic under multiple Virtual Channels (VCs) with mixed Reliable Transfer (RT) and Continuous Transfer (CT) modes, particularly in handling packet retransmission and ensuring quality of service, especially when dealing with large packet sizes and varying traffic patterns.
Innovation Solution
Implementing a unified final buffer with a page-based and pointer-based scheme, allowing for dynamic buffer management and flexible allocation of memory pages across VCs, which enables prioritization and efficient handling of packets through Virtual Output Queues (VOQs) and Retransmission Mappers, preventing head-of-line blocking and optimizing throughput.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a unified final buffer with page-based and pointer-based scheme is implemented, then traffic optimization and throughput are improved, but device complexity increases
Solution Approach 1:
The final buffer is segmented into multiple pages, with each page independently manageable. The buffer memory is divided into first and second portions for different VC types, and each portion is further divided into multiple pages. This segmentation allows flexible allocation and management of buffer space for different traffic types while maintaining overall system throughput.
Solution Approach 2:
The unified final buffer structure serves multiple functions: it handles both RT mode packets (requiring retransmission) and CT mode packets (loss-tolerant) simultaneously, manages multiple VCs with different QoS requirements, and provides both page-based allocation for RT mode and pointer-based allocation for CT mode. This multi-functionality improves productivity without requiring separate buffer structures for each traffic type.
2Loss of time
If dynamic buffer management with VOQs and Retransmission Mappers is implemented, then packet prioritization and latency are improved, but device complexity increases
Solution Approach 1:
Virtual Output Queues (VOQs) are pre-established for different VC types and priorities before packet arrival. The Retransmission Mapper is pre-configured with mappings between packet identifiers and their corresponding buffer locations. This preliminary structure enables immediate packet prioritization and routing decisions, reducing latency without requiring complex real-time computations.
Solution Approach 2:
The Retransmission Mapper acts as an intermediary structure that bridges the packet input and buffer allocation. It stores the relationship between packet identifiers and buffer page locations, enabling efficient retrieval and retransmission of packets without direct complex interactions between multiple buffer management components. This intermediary simplifies the overall control logic while maintaining low latency.
3Reliability
If separate buffer management for RT and CT mode VCs is avoided, then device complexity is reduced, but reliability of RT mode transfer deteriorates
Solution Approach 1:
Different buffer allocation strategies are applied locally to different VC types within the same unified buffer structure. RT mode VCs receive page-based allocation with guaranteed buffer pages for reliable retransmission, while CT mode VCs use pointer-based allocation for maximum throughput. This local differentiation ensures RT mode reliability without requiring completely separate buffer structures, thus avoiding excessive complexity.
4Reliability
If maximum buffer space is allocated for retransmission, then reliability of RT mode is improved, but throughput for CT mode packets deteriorates
Solution Approach 1:
The buffer allocation is dynamic rather than static. Buffer pages are allocated and released based on actual packet arrival and transmission status. For RT mode packets, buffer pages are held until successful delivery or retransmission completion. For CT mode packets, buffer space is released immediately after transmission. This dynamic allocation ensures RT mode reliability while maximizing CT mode throughput by preventing unnecessary buffer occupancy.
Data Source
AI summary
A method and apparatus for unified final buffer with pointer-based and page-based scheme for traffic optimization have been disclosed.