Shared Packet Memory System for Ethernet Switch Incast Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Ethernet switches face throughput bottlenecks in incast applications due to limited bandwidth utilization and inefficient memory management, particularly in handling simultaneous bursts of packets, which leads to packet drops and wasted bandwidth.
Innovation Solution
Implementing a shared packet memory system with a queueing structure that combines direct communication between input and output ports using a shared memory architecture, where tags are used to manage packet data and metadata, reducing the number of output queue slices and hiding latency through a tag distribution bus and queue cache structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional separate queue structures are used for each output port, then packet management is straightforward, but the number of queues increases significantly and memory efficiency decreases
Solution Approach 1:
The patent combines multiple separate output queues into a single shared queue structure that serves all output ports. Instead of maintaining N separate queues for N output ports, a single shared queue manages packets for all ports, significantly reducing the number of queue structures and improving memory utilization efficiency.
Solution Approach 2:
The shared queue structure serves multiple output ports simultaneously, making a single data structure universal for all egress operations. This multi-functional approach allows the same queue to manage packets destined for different output ports, eliminating the need for dedicated queues for each port.
2Productivity
If packets are fetched from switch fabric to egress port, then packet transmission can proceed, but bandwidth is wasted when packets are dropped after fetching
Solution Approach 1:
The egress pipeline performs preliminary actions by examining packet headers and metadata before packets are fully fetched from the switch fabric. Packets that fail policy checks are identified and dropped in advance, preventing wasted bandwidth from transmitting packets that will ultimately be discarded.
Solution Approach 2:
The egress pipeline acts as an intermediary between the switch fabric and the egress port. It inspects packets from the switch fabric, applies policy decisions based on headers and metadata, and only permits compliant packets to reach the egress port, thereby preventing bandwidth waste from transmitting dropped packets.
3Reliability
If more output queue slices are created to handle incast bursts, then packet loss during bursts is reduced, but memory overhead and system complexity increase
Solution Approach 1:
The shared queue structure dynamically adapts to handle incast bursts by allowing flexible allocation and deallocation of buffer space based on real-time traffic conditions. Instead of static queue slices that increase complexity during bursts, the dynamic shared queue automatically adjusts to accommodate burst traffic while maintaining packet delivery reliability.
Data Source
AI summary
Examples describe use of multiple meta-data delivery schemes to provide tags that describe packets to an egress port group. A tag, that is smaller than a packet, can be associated with a packet. The tag can be stored in a memory, as a group with other tags, and the tag can be delivered to a queue associated with an egress port. Packets received at an ingress port can be as non-interleaved to reduce underrun and providing cut-through to an egress port. A shared memory can be allocated to store packets received at a single ingress port or shared to store packets from multiple ingress ports.