DPU Packet Processing With Watermark-Based Queue Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing DPU packet processing methods fail to effectively manage network traffic bursts, leading to head-of-line blocking due to limited buffer space and inefficient handling of packets requiring different processing speeds.

Innovation Solution

Implement a packet processing method that monitors buffer watermarks to prioritize packets requiring longer processing times, discarding them when the buffer threshold is exceeded, thereby reducing the impact on faster paths and avoiding buffer overflow.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If packets are stored in a shared buffer unit for processing, then buffer space utilization is improved, but head-of-line blocking occurs when slow packets occupy buffer space needed by fast packets

Engineering Contradiction:
Improvebuffer space utilizationVSAvoidpacket processing efficiency
Core Design Contradiction:
Quantity of substanceVSProductivity

Solution Approach 1:

The patent segments the buffer management by creating separate queue structures (first queue for slow packets, second queue for fast packets) within the shared buffer unit. This segmentation allows independent management of packets requiring different processing speeds, preventing slow packets from blocking fast packet processing while maintaining overall buffer space utilization.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by implementing different buffer management strategies for different packet types. The first queue uses a drop-tail policy for slow packets, while the second queue uses a priority policy for fast packets. This localized differentiation resolves the contradiction by allowing each queue to be optimized for its specific packet processing requirements.

Inventive Principle:
Principle #3Local quality

2Adaptability or versatility

If buffer space is allocated for slow processing packets, then comprehensive packet handling is improved, but the bandwidth of fast paths is reduced due to head-of-line blocking

Engineering Contradiction:
Improvepacket handling capabilityVSAvoidfast path bandwidth
Core Design Contradiction:
Adaptability or versatilityVSSpeed

Solution Approach 1:

The patent segments fast and slow packets into separate queues (second queue for fast packets, first queue for slow packets) with independent processing paths. This segmentation ensures that slow packet processing does not block fast packet transmission, maintaining high fast path bandwidth while preserving comprehensive packet handling capability through the first transmission path.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary mechanism (the buffer management unit with watermark monitoring) that mediates between slow and fast packet processing. When buffer watermark exceeds thresholds, the system selectively discards slow packets from the first queue while preserving fast packets in the second queue, thus protecting fast path bandwidth while maintaining adaptability.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Device complexity

If all packets are processed through the same transmission path, then processing simplicity is maintained, but overall network performance deteriorates due to slow packets bottlenecking the system

Engineering Contradiction:
Improvetransmission path structureVSAvoidnetwork packet processing performance
Core Design Contradiction:
Device complexityVSProductivity

Solution Approach 1:

The patent segments the transmission paths into a first transmission path for slow packets and a second transmission path for fast packets, with corresponding separate queues in the buffer unit. This segmentation enables parallel processing of different packet types at different speeds, significantly improving overall network performance while maintaining relatively simple path structures through centralized buffer management.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentUS20250310277A1Packet processing method and apparatus applied to DPU, computer device and storage medium
Publication Date: 2025.10.02 BEIJING VOLCANO ENGINE TECH CO LTD
  • US20250310277A1 patent drawing
  • US20250310277A1 patent drawing
  • US20250310277A1 patent drawing

AI summary

The present disclosure relates to the field of cloud computing technology, and discloses a packet processing method and apparatus applied to a DPU, a computer device, and a storage medium. The packet processing method includes: storing a received network packet in a buffer unit, where the network packet includes a first packet and a second packet, and a packet processing speed of the second transmission path is higher than a packet processing speed of the first transmission path; monitoring a buffer watermark of a packet buffer space of the buffer unit, where the buffer watermark is used to indicate a usage degree of the packet buffer space; and in response to the buffer watermark being greater than a preset threshold, sending the first packet in the buffer unit to the first transmission path.