RAID Sub-Stripe Placement With Multi-Shard Failure Domain Protection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing storage systems face inefficiencies in managing data storage and retrieval across multiple drives, particularly in handling failures and optimizing operations without redundant write processes.
Innovation Solution
Implementing a direct-mapped flash storage system that addresses data blocks directly without translation by storage controllers, utilizing non-volatile RAM for buffering, and offloading device management tasks from storage drives to controllers, enabling higher-level system operations to manage zones and allocation units across multiple drives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data is striped across multiple drives with traditional storage controllers, then data management and protection functions are provided, but unnecessary write operations occur and latency increases
Solution Approach 1:
The patent extracts the translation and management functions from traditional storage controllers and relocates them to the host system. The storage drives directly map data blocks without controller intervention, eliminating unnecessary write operations and reducing latency while maintaining data protection through host-managed redundancy
Solution Approach 2:
The patent introduces a host-based data management layer that acts as an intermediary between the application and storage drives. This layer handles data striping, redundancy, and failure domain management, allowing drives to operate in direct-mapped mode without complex controller translation, thus reducing write latency while maintaining reliability
2Ease of operation
If storage controllers manage data blocks and translation, then data management is centralized, but device complexity and operation overhead increase
Solution Approach 1:
The patent extracts complex translation and management logic from storage controllers and moves it to the host system. Drives are simplified to perform only direct block mapping, while the host handles striping, redundancy, and failure domain management, reducing device complexity without sacrificing ease of operation
Solution Approach 2:
The patent inverts the traditional architecture by having the host (rather than the controller) perform data management functions. The host manages zones, allocation units, and data striping across failure domains, while drives simply execute direct mappings, thereby reducing controller complexity while maintaining centralized management capabilities
3Reliability
If multiple shards are placed in separate failure domains, then data protection is improved, but storage efficiency and space utilization decrease
Solution Approach 1:
The patent segments data into allocation units that can be strategically placed across failure domains. Multiple shards of the same allocation unit can reside within a single failure domain while maintaining protection through redundancy mechanisms, improving storage efficiency without compromising data protection
Solution Approach 2:
The patent changes the parameter of shard distribution from strict one-shard-per-failure-domain to flexible multi-shard-per-failure-domain placement. By adjusting the placement strategy and using host-managed redundancy, the system achieves both high storage efficiency and data protection
Data Source
AI summary
Redundant array of independent drives (RAID) sub-stripes are formed across one or more solid-state storage devices of storage nodes of a storage system. The RAID sub-stripes include corresponding shards of data to be stored at the solid-state storage devices, wherein at least one of the RAID sub-stripes has at least two of the corresponding shards of data on a same storage node. At least one global parity shard is generated for the RAID sub-stripes. The at least one global parity shard is to be stored on a first storage node that is different from each of the same storage nodes storing the at least two of the corresponding shards of data.