RDMA-to-NVMe Request Conversion for Low-Latency NOF Storage

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

There is a significant access delay in accessing storage disks in Non-Volatile Memory Express over Fabric (NOF) systems, leading to low input/output operations per second (IOPS) capabilities in NOF storage networks.

Innovation Solution

The method involves converting access requests from a remote direct memory access (RDMA) format directly into an NVMe format within the storage apparatus, bypassing complex block device layer processing, thereby shortening the transmission path and reducing delays.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If access requests are processed through the block device layer, then comprehensive data management and control are achieved, but transmission delay increases and IOPS capability decreases

Engineering Contradiction:

Improvedata management controlVSAvoidaccess delay

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent extracts the access request processing path from the traditional block device layer and creates a direct RDMA to NVMe conversion path. This removes unnecessary processing steps while maintaining essential data management functions, thereby reducing access delay without completely eliminating control mechanisms.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent segments the storage access path into distinct functional components: RDMA request reception, format conversion, NVMe queue management, and persistent storage access. This segmentation allows optimization of each segment independently, particularly by streamlining the conversion process and reducing overhead in the critical path.

Inventive Principle:

Principle #1Segmentation

2Reliability

If complex block device layer processing is performed, then thorough request validation and management are ensured, but transmission path lengthens and processing speed decreases

Engineering Contradiction:

Improverequest validationVSAvoidprocessing speed

Core Design Contradiction:

ReliabilityVSSpeed

Solution Approach 1:

The patent performs preliminary validation and setup actions before the access request reaches the conversion stage. This includes pre-configuring NVMe queues, validating RDMA request parameters upfront, and establishing mapping relationships in advance, so that the actual conversion and execution paths can proceed with minimal additional validation overhead.

Inventive Principle:

Principle #10Preliminary action

3Reliability

If BIO resources are allocated for each access request, then resource management and tracking are improved, but resource consumption increases and system overhead grows

Engineering Contradiction:

Improveresource managementVSAvoidresource consumption

Core Design Contradiction:

ReliabilityVSLoss of energy

Solution Approach 1:

The patent makes BIO resources universal by allowing a single BIO resource to serve multiple access requests through the shared NVMe queue infrastructure. Instead of allocating dedicated BIO resources for each request, the system uses a pool of BIO resources that can be dynamically assigned and reused, reducing overall resource consumption while maintaining proper tracking and management through the NVMe completion queues.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12535949B2Method and apparatus for processing access request, storage apparatus, and storage medium

Publication Date: 2026.01.27 HUAWEI TECH CO LTD

AI summary

An access request in a remote direct memory access (RDMA) format is first obtained from the internal memory of the storage apparatus, where the access request is used for requesting to access the persistent storage medium in the storage apparatus. Then, the access request in the RDMA format is directly converted into an access request in a Non-Volatile Memory Express (NVMe) format, and a target operation corresponding to the access request in the NVMe format is executed to access the persistent storage medium.