Selective Packet Retransmission in RDMA NICs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
RDMA protocols face inefficiencies in packet retransmission due to lossless network designs, leading to unnecessary duplication of packets and increased network load, especially when significant packet loss occurs.
Innovation Solution
Implementing selective acknowledgment and retransmission mechanisms in network interface controllers (NICs) that differentiate retransmission modes based on transaction types, such as RDMA Write, Read, and Send, allowing only missing packets to be retransmitted in certain cases, reducing the burden on receiving NICs and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all packets are retransmitted upon receiving a NAK packet, then reliability is improved, but network load increases and efficiency deteriorates
Solution Approach 1:
The patent segments the retransmission process by identifying and retransmitting only the specific missing packet(s) indicated in the NAK packet, rather than retransmitting all packets. The NAK packet contains a packet serial number identifying the missing packet, allowing the transmitter to selectively retransmit only that specific packet segment.
Solution Approach 2:
The patent applies different retransmission behaviors based on the type of RDMA transaction. For Write and Read transactions, only the missing packet is retransmitted, while for Send transactions, all subsequent packets are retransmitted. This local differentiation optimizes reliability and efficiency for each transaction type specifically.
2Loss of energy
If selective retransmission is implemented, then network load is reduced, but device complexity increases
Solution Approach 1:
The patent implements the selective retransmission logic directly in the NIC hardware, allowing the NIC to autonomously handle packet retransmission decisions without involving the host processor or operating system. The NIC independently processes NAK packets, identifies missing packets, and retransmits them, making the system self-sufficient for error recovery.
Solution Approach 2:
The patent introduces a dedicated retransmission buffer in the NIC that temporarily holds transmitted packets awaiting acknowledgment. This intermediary buffer allows the NIC to efficiently manage retransmissions by having packets readily available for selective retransmission without requiring complex host processor intervention.
3Speed
If host processor involvement is eliminated, then processing speed improves, but control precision may worsen
Solution Approach 1:
The patent dynamically adjusts retransmission behavior based on the type of RDMA transaction being performed. The NIC examines the transaction type (Write, Read, or Send) and automatically selects the appropriate retransmission strategy: selective retransmission for Write/Read transactions or full retransmission for Send transactions. This dynamic adaptation enables the system to optimize for both speed and precision based on transaction requirements.
Data Source
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.