Four-Controller Storage Array With Cache Mirroring Fault Isolation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-controller storage arrays, both loosely and tightly coupled, suffer from low availability due to simultaneous failures of controller pairs or shared fault domains, leading to inaccessible hard disks and lack of cache redundancy.
Innovation Solution
A multi-controller storage array design with four-controller modules connected via Ethernet, utilizing cache mirroring and RDMA networks like RoCE, allowing each node to access shared storage expansion disk enclosures and ensuring cache redundancy through target mirror nodes, even in the event of failures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If controller nodes are tightly coupled and share backend hard disks, then processing performance improves, but availability deteriorates when a backplane fails
Solution Approach 1:
The storage array is divided into multiple independent controller groups, where each group contains controller nodes that share a backend hard disk. These groups are connected via Ethernet network, creating separate fault domains. When a backplane fails in one group, only controllers in that group are affected, while other groups continue to provide services, thus resolving the contradiction between tight coupling for performance and fault isolation for availability.
2Reliability
If controller nodes are loosely coupled with separate backend hard disks, then fault isolation improves, but storage resource utilization deteriorates
Solution Approach 1:
Backend hard disks are designed to be shareable across multiple controller groups through the Ethernet network connection. A single backend hard disk can be accessed by controllers from different groups, enabling resource pooling and improved utilization. This maintains fault isolation between groups while allowing efficient resource sharing, resolving the contradiction between fault isolation and resource utilization.
3Speed
If cache is localized to individual controller nodes, then access speed improves, but cache redundancy deteriorates when a controller node fails
Solution Approach 1:
The patent implements cache mirroring where cache data is replicated across multiple controller nodes within the same controller group. When a controller node fails, its cached data can be retrieved from the mirrored cache in another node within the group. This maintains fast local cache access while providing redundancy through copying, resolving the contradiction between access speed and cache redundancy.
Data Source
AI summary
Embodiments of the present application provide a multi-controller storage array, a storage system, a data processing method, and a storage medium. The storage array includes a four-controller storage module, and the four-controller storage module includes two controller pairs and a storage expansion disk enclosure. The two controller pairs are connected via Ethernet. Any controller pair of the two controller pairs includes two controller nodes. Each controller node is connected to the storage expansion disk enclosure for shared access to the storage expansion disk enclosure. Each controller node is provided with a cache unit, and any controller node, in response to a write data request, determines a target mirror controller node corresponding to the controller node, wherein a cache mirror of the controller node is stored in a cache unit of the target mirror controller node.