Storage Controller Data Replication via Direct Memory Access
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Software Defined Storage (SDS) systems face high processing and memory consumption, as well as Input/Output (I/O) latency due to data replication across servers, which traverses the entire storage stack, leading to inefficiencies in data redundancy.
Innovation Solution
A redundant storage system is implemented with direct memory access (DMA) capabilities between storage controllers, allowing data to bypass the host's processing and memory, thereby reducing latency and resource consumption by mapping memory directly between storage controllers and hosts, enabling efficient data replication without traversing the storage stack.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data replication traverses the entire storage stack through host systems and networks, then data redundancy is achieved, but processing and memory consumption increase significantly
Solution Approach 1:
The patent extracts the data replication function from the host system's storage stack and relocates it to a dedicated storage controller. The storage controller independently handles data replication between storage devices, removing the processing and memory burden from the host system while maintaining data redundancy functionality.
Solution Approach 2:
The storage controller acts as an intermediary between the host system and storage devices. It intercepts data I/O operations and performs replication in the storage layer without requiring host system involvement, thus reducing processing and memory consumption on the host while achieving data redundancy.
2Reliability
If data replication traverses the entire storage stack through host systems and networks, then data redundancy is achieved, but I/O latency increases
Solution Approach 1:
The patent extracts the replication function from the host storage stack and places it in the storage controller, enabling local replication without network traversal. This eliminates the time-consuming network communication and OS-level processing, significantly reducing I/O latency while maintaining data redundancy.
Solution Approach 2:
The patent segments the storage system into distinct layers: the host system handles high-level I/O requests, while the storage controller handles low-level replication operations. This segmentation allows replication to occur independently at the storage layer without blocking host operations, reducing overall I/O latency.
3Productivity
If memory is mapped directly between storage controllers and hosts, then data transfer efficiency improves, but system complexity increases
Solution Approach 1:
The storage controller performs self-service by autonomously managing memory mapping and data replication without requiring complex host system configuration. The controller handles memory allocation and mapping internally, simplifying the overall system configuration while maintaining high data transfer efficiency.
Data Source
AI summary
Embodiments herein provide for redundant data storage. One storage system includes first and second host systems each comprising a memory and a persistent storage device. The storage system also includes first and second storage controllers each comprising a memory (e.g., DRAM). The memory of the first storage controller is mapped to the memory of the first host system and the memory of the second storage controller is mapped to the memory of the second host system. The first storage controller is operable to DMA data from the persistent storage device of the first host system to the memory of the first storage controller, and to direct the second storage controller to DMA the data to the persistent storage device of the second host system via the memory of the second storage controller.