Host-Aware Update Write Atomicity in Erasure Coded Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Software-defined storage systems implementing erasure coding for data protection face challenges in maintaining data integrity during update write operations, particularly in the absence of non-volatile storage, leading to increased I/O load and reduced throughput due to the need for write-ahead logging, which results in data integrity exposures and inefficiencies.
Innovation Solution
A host-aware update write protocol is implemented, which maintains mapping data between logical and physical addresses, allowing for atomic multi-device operations by tracking prior version copies and using a copy-on-write methodology to reduce write operations and ensure data integrity through prior version mapping information stored on the storage unit.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If software-defined storage systems implement erasure coding for data protection, then data integrity is improved, but I/O load increases and throughput decreases due to write-ahead logging requirements
Solution Approach 1:
The patent extracts the write-ahead logging function from the storage system architecture by implementing host-aware update writes that perform atomic multi-device operations directly. This eliminates the need for separate write-ahead log structures, removing the I/O overhead while maintaining data integrity through the host's responsibility to coordinate atomic updates across multiple devices.
Solution Approach 2:
The patent applies preliminary action by having the host system prepare and execute atomic update operations that simultaneously modify multiple storage devices. By performing the coordination and atomicity guarantees at the host level before writes occur, the system avoids the need for post-write verification and recovery mechanisms, reducing I/O load while ensuring data integrity.
2Reliability
If write-ahead logs are implemented for atomic read-modify-write operations, then data protection is improved, but I/O load increases and response time increases
Solution Approach 1:
The patent merges the write-ahead logging function with the host's update operation by implementing atomic multi-device writes at the host level. This consolidation eliminates the need for separate log write and recovery operations, reducing the number of I/O operations and response time while maintaining data protection through the atomicity guarantee provided by the host-aware protocol.
Solution Approach 2:
The patent enables self-service by making the host system responsible for coordinating atomic update operations across multiple storage devices. The host uses mapping data from the storage unit to identify current and prior physical addresses, then performs atomic updates without requiring the storage system to provide write-ahead logging services, thereby reducing system-level I/O load and response time.
3Reliability
If additional storage devices are used for write-ahead logs, then data integrity is improved, but device complexity and I/O load increase
Solution Approach 1:
The patent applies universality by making the host system perform multiple functions: it manages data updates, coordinates atomic operations across multiple devices, and maintains awareness of storage mapping data. This multi-functionality at the host level replaces the need for dedicated write-ahead log storage devices, reducing device complexity while maintaining data integrity through the host's coordinated atomic updates.
Solution Approach 2:
The patent uses the host system as an intermediary that coordinates between the application layer and the storage system. The host uses mapping data from the storage unit to identify physical addresses and performs atomic updates, acting as a mediator that ensures data integrity without requiring additional storage devices for write-ahead logs, thereby reducing system complexity.
4Reliability
If mapping data between logical and physical addresses is maintained, then atomic multi-device operations are enabled, but storage unit complexity increases
Solution Approach 1:
The patent applies partial action by having the storage unit maintain only the essential mapping data between logical block addresses and physical block addresses needed for atomic operations, rather than maintaining complete metadata about all data operations. This selective maintenance of mapping information enables atomic multi-device operations while minimizing the complexity burden on the storage unit.
Data Source
AI summary
One embodiment provides a method comprising maintaining, on a storage unit, mapping data between a first set of logical addresses (e.g., logical block addresses or LBAs) viewed by a host and a first set of physical addresses (e.g., physical block addresses or PBAs) and a second set of physical addresses of the storage unit. A first logical address (e.g., LBA) of the first set of logical addresses corresponds to a first physical address (e.g., PBA) of the first set of physical addresses that maintains current data for the first logical address. The first logical address further corresponds to a second physical address (e.g., PBA) of the second set of physical addresses that maintains prior data for the first logical address. The method further comprises receiving, at the storage unit, a command from the host to perform a multi-device operation involving the first logical address. The operation is performed atomically.