RDMA Selective Packet Retransmission for Out-of-Order Loss Recovery

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

RDMA protocols face inefficiencies in packet retransmission due to lossless networks, leading to duplicate transmission of many packets upon single packet loss, which negatively impacts performance when significant loss events occur.

Innovation Solution

Implementing selective retransmission mechanisms by NICs that handle packet loss based on transaction type, retransmitting only missing packets for certain RDMA transactions and all subsequent packets for others, reducing the burden on receiving NICs and optimizing recovery from packet loss.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional RDMA protocols retransmit all subsequent packets upon packet loss, then reliability is maintained, but network performance deteriorates due to duplicate transmissions

Engineering Contradiction:
Improvepacket delivery reliabilityVSAvoidnetwork performance
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent segments the retransmission process by identifying and retransmitting only the specific lost packet rather than all subsequent packets. The NAK packet mechanism allows the receiver to precisely indicate which packet is missing, enabling the transmitter to send only that specific packet for retransmission, thus resolving the contradiction between reliability and network performance.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by treating different packets differently based on their reception status. Instead of uniformly retransmitting all packets after a loss event, the system selectively retransmits only the specific lost packet identified by the NAK mechanism, optimizing resource usage while maintaining reliability.

Inventive Principle:
Principle #3Local quality

2Productivity

If selective retransmission is implemented, then network performance improves by reducing duplicate transmissions, but device complexity increases due to NAK packet handling

Engineering Contradiction:
Improvenetwork performanceVSAvoidNIC complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent extracts the acknowledgment functionality from traditional ACK-based protocols and implements a dedicated NAK (Negative Acknowledgment) mechanism. This separate NAK path allows for precise identification of lost packets without the complexity of full selective acknowledgment, improving network performance while keeping device complexity manageable.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The NAK mechanism enables the receiver to self-identify lost packets and request specific retransmissions without requiring complex transmitter-side packet tracking. The receiver autonomously determines which packet is missing and sends a targeted NAK, reducing the computational burden on both ends.

Inventive Principle:
Principle #25Self-service

3Device complexity

If all packets are retransmitted upon loss detection, then simplicity of protocol is maintained, but loss of time increases due to unnecessary retransmissions

Engineering Contradiction:
Improveprotocol simplicityVSAvoidretransmission time
Core Design Contradiction:
Device complexityVSLoss of time

Solution Approach 1:

The patent implements preliminary action by having the receiver prepare and send NAK packets immediately upon detecting out-of-order arrivals, rather than waiting for timeout periods. This proactive approach identifies lost packets faster and initiates selective retransmission, reducing the time penalty compared to traditional methods.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4002734B1Selective retransmission of packets
Publication Date: 2026.02.25 MELLANOX TECHNOLOGIES LTD(IL)
  • EP4002734B1 patent drawingFigure 1
  • EP4002734B1 patent drawingFigure 2
  • EP4002734B1 patent drawingFigure 3

AI summary

A method for data transfer includes transmitting a sequence of data packets from a first computer over a network to a second computer in a single RDMA data transfer transaction. Upon receipt of a second packet in the sequence without previously having received the first packet, the second computer sends a NAK packet over the network to the first computer, indicating that the first packet was not received. A retransmission mode is selected responsively to the type of the transaction, such that when the transaction is of a first type, the first packet is retransmitted from the first computer to the second computer in response to the NAK packet without retransmitting the second packet, and when the transaction is of a second type, both the first and second packets are retransmitted from the first computer to the second computer in response to the NAK packet.