Shared Packet Buffer Isolation With Dynamic Threshold Admission
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network transmission systems face challenges in managing heterogeneous traffic due to bursty nature, leading to temporary congestion, packet loss, and increased latency, with current solutions either over-allocating resources or providing shared buffers that result in inefficient utilization and unfair consumption of buffer memory.
Innovation Solution
A two-tier admission control scheme is implemented, where each type of traffic flow is allocated a guaranteed minimum buffer space, followed by shared buffer space using dynamic thresholds, ensuring fair access and minimizing memory wastage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If resources are optimized for average utilization, then network fabric utilization is reasonable on average, but severe temporary congestion occurs during traffic bursts leading to packet loss
Solution Approach 1:
The shared packet buffer is segmented into multiple isolated regions, each dedicated to a specific traffic type (network-network, network-host, host-network, host-host). This segmentation allows each region to independently handle its traffic bursts without causing congestion in other regions, thereby preventing packet loss while maintaining high overall utilization.
Solution Approach 2:
Each traffic type is assigned a minimum guaranteed space with specific admission control thresholds tailored to its characteristics. This local quality customization allows each traffic category to have appropriate buffer management parameters, ensuring reliable handling of bursty traffic for each type while optimizing overall buffer utilization.
2Productivity
If a shared buffer is used for all traffic types, then memory utilization is high, but fair access is not provided and memory wastage occurs
Solution Approach 1:
The buffer is divided into isolated regions for different traffic types, with each region having its own admission control mechanism. This segmentation ensures that one traffic type cannot monopolize the buffer, providing fair access while maintaining high overall utilization through dynamic threshold adjustments.
Solution Approach 2:
Dynamic admission control thresholds are implemented for each traffic type, allowing the buffer allocation to adapt in real-time to changing traffic conditions. This dynamic approach ensures fair access by preventing any single traffic type from dominating the buffer while maximizing memory utilization through flexible resource allocation.
3Reliability
If admission control is implemented to prevent buffer overflow, then packet loss is reduced, but throughput is reduced due to conservative buffer allocation
Solution Approach 1:
By segmenting the buffer into isolated regions and implementing admission control at the region level rather than using a single conservative global threshold, the system can allow higher throughput for each traffic type within its allocated space while still preventing buffer overflow and packet loss.
Solution Approach 2:
Admission control thresholds are dynamically adjusted based on buffer occupancy levels and traffic characteristics. This parameter change approach allows the system to maintain reliable packet loss prevention while optimizing throughput by adapting buffer allocation parameters in real-time rather than using fixed conservative values.
Data Source
AI summary
A system and method for managing shared memory packet buffers is disclosed. In some embodiments, the system is configured to receive and classify a packet as one of: network-network, network-host, host-network, or host-host; select a minimum guarantee space for the packet according to the classification thereof; if the selected minimum guarantee space is available, store the packet therein; otherwise, if a dedicated shared space is available, store the packet therein; otherwise, if a global shared space is available, store the packet therein; and otherwise, drop the packet.