NVMeoF Burst Mode RDMA Write Packet Transmission

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The transmission of Non-Volatile Memory Express over Fabric (NVMeoF) commands from a host to a target device can result in significant transport bandwidth overhead and excessive hardware resource requirements, particularly when using Remote Direct Memory Access (RDMA) networks.

Innovation Solution

Implementing a burst mode transfer method using RDMA Write packets for submitting NVMeoF commands and responses between a host and a target device, where commands are initiated and responses are posted in pre-registered memory regions of Submission and Completion Queues, respectively, to reduce overhead and resource demands.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If NVMeoF commands are transmitted using traditional RDMA methods, then reliable storage access is achieved, but transport bandwidth overhead increases and hardware resource requirements become excessive

Engineering Contradiction:
Improvestorage access reliabilityVSAvoidtransport bandwidth overhead
Core Design Contradiction:
ReliabilityVSLoss of energy

Solution Approach 1:

The patent combines multiple NVMeoF commands into a single RDMA Write packet, allowing batched transmission of multiple commands from host to target device. This merging approach reduces the number of separate RDMA packets required, thereby decreasing transport bandwidth overhead while maintaining reliable storage access through the established NVMeoF protocol framework

Inventive Principle:
Principle #5Merging (Combining)

2Adaptability or versatility

If NVMeoF commands are transmitted using traditional RDMA methods, then storage access functionality is maintained, but hardware resource requirements in the target device become excessive

Engineering Contradiction:
Improvestorage access functionalityVSAvoidhardware resource requirements
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent merges multiple command submissions into a single RDMA Write packet that targets a pre-registered memory region containing a Submission Queue. This approach reduces the number of individual packet processing operations required in the target device, thereby decreasing hardware resource requirements while preserving full NVMeoF storage access functionality through the standardized queue-based interface

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentUS11016911B2Non-volatile memory express over fabric messages between a host and a target using a burst mode
Publication Date: 2021.05.25 SAMSUNG ELECTRONICS CO LTD
  • US11016911B2 patent drawing
  • US11016911B2 patent drawing
  • US11016911B2 patent drawing

AI summary

Systems and methods for managing transfer of NVMeoF commands/responses between a host and a target are described. The systems and methods may initiate and convert at least one Input/Output request into at least one Non-Volatile Memory Express over Fabric (NVMeoF) command to access a storage device attached with the target device. A host may transmit the at least one NVMeoF command in a burst mode using a Remote Direct Memory Access (RDMA) Write packet to a pre-registered memory region of the target device. In response to reception of the at least one NVMeoF command, the target device may post at least one NVMeoF completion response corresponding to the at least one NVMeoF command using the RDMA Write packet to a pre-registered memory region of the host.