DPU Emulated Storage for NVMe Over Fabrics DMA Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current networked storage systems require multiple direct memory access (DMA) transfers to access local storage devices, leading to inefficiencies and performance bottlenecks, especially when using SmartNICs and NVMe over Fabrics technology.
Innovation Solution
The implementation of an emulated storage device using DPU software that allows a host to access local storage devices with fewer than three DMA transfers by utilizing a fabric target service to recognize and process NVMe requests, enabling direct memory access between the host's application buffer and the local NVMe device without network data movement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If networked storage systems use multiple DMA transfers to access local storage devices, then storage protocols can be supported (NAS and SAN), but performance bottlenecks and inefficiencies occur
Solution Approach 1:
The patent introduces a DPU as an intermediary device between the host and storage devices. The DPU intercepts storage I/O requests from the host and directly accesses local storage devices, bypassing the traditional networked storage path. This intermediary approach allows the system to maintain protocol support capabilities while achieving direct storage access performance, effectively resolving the contradiction between versatility and productivity.
2Adaptability or versatility
If SmartNICs are used for storage access, then networked storage functionality is enabled, but multiple DMA transfers are required causing inefficiency
Solution Approach 1:
The patent extracts the storage access functionality from the networked storage path and redirects it through the DPU. By taking out the storage I/O requests from the traditional NAS/SAN path, the system can handle them through a more efficient direct path while still maintaining networked storage capabilities for other scenarios. This extraction reduces the number of DMA transfers required while preserving functionality.
3Adaptability or versatility
If data is accessed via networked storage protocols, then distributed file systems and block storage are supported, but data access latency increases
Solution Approach 1:
The patent implements preliminary action by having the DPU intercept and handle storage requests before they proceed through the full networked storage stack. The DPU is pre-configured with knowledge of local storage devices and can directly route requests to them, bypassing unnecessary network protocol processing. This preliminary intervention reduces data access latency while maintaining support for distributed storage architectures.
Data Source
AI summary
A computerized system operating in conjunction with computerized apparatus and with a fabric target service in data communication with the computerized apparatus, the system comprising functionality residing on the computerized apparatus, and functionality residing on the fabric target service, which, when operating in combination, enable the computerized apparatus to coordinate access to data.