Distributed Storage Write Response Latency Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing distributed storage systems experience increased write operation delays due to the need for multiple responses from storage nodes before a write completion response can be sent to the host, affecting write performance.
Innovation Solution
The proposed solution involves a distributed storage system where the client generates data strips using an erasure coding algorithm and sends them to storage nodes, allowing for a write success response to be sent before the write completion response, thereby reducing interactions and delays.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the storage node sends two responses to the client before the client sends the write completion response to the host, then atomicity of the write operation is ensured, but the write operation delay is increased
Solution Approach 1:
The storage node pre-establishes the mapping between host access address and strip identifier in the write log during the preparation phase, before the commit phase is executed. This preliminary action allows the client to send the write completion response to the host immediately after the commit phase succeeds, without waiting for the storage node to convert write log content to accessible metadata and send additional responses.
2Reliability
If the client waits for multiple responses from storage nodes before sending write completion response to host, then data consistency is ensured, but write performance is affected
Solution Approach 1:
The patent extracts the write completion response sending action from the traditional two-phase commit process. By pre-establishing mappings in the write log during preparation phase and using asynchronous metadata conversion, the client can send the write completion response to the host independently and immediately after commit phase success, without being blocked by waiting for storage node responses about metadata conversion status.
Data Source
AI summary
A distributed storage system comprises a data storage node Nx that receives a write success response returned by a hard disk that provides the storage address. The data storage node Nx returns a write success response to a client. The client receives the write success response returned by the data storage node Nx, and the client returns a write completion response to a host.