Distributed Storage Write Response Latency Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Distributed storage systems face deteriorated write response performance due to the need for data redundancy, which involves duplicating updated data across multiple nodes with SSDs, leading to increased latency and potential data loss in case of power interruptions.
Innovation Solution
A distributed storage system configures a first computer to cache updated data and select n second computers with different power supply units as transfer destinations, allowing the first computer to return a write response without waiting for all n computers to cache the data, thereby reducing write response time and preventing data loss even in case of power failures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If updated data is duplicated to multiple nodes with SSD as cache to ensure data redundancy, then data protection is improved, but write response performance deteriorates due to waiting for all nodes to cache data
Solution Approach 1:
The patent applies partial action by selecting only n computers with different power supply units as transfer destinations instead of duplicating data to all available nodes. This partial replication provides sufficient data protection while reducing the waiting time for write operations, as the system only needs to ensure data reaches computers with different power supplies rather than all possible destinations.
Solution Approach 2:
The system performs preliminary action by pre-selecting n computers with different power supply units before the write operation occurs. This advance selection of transfer destinations with diverse power sources enables the system to quickly identify valid targets without waiting during the write process, thus improving write response performance while maintaining data protection.
2Reliability
If data is written to nonvolatile storage device (SSD) before responding to client to ensure data permanence, then data reliability is improved, but write response performance deteriorates
Solution Approach 1:
The patent uses copying by creating replicas of updated data in cache memory of selected nodes with different power supply units. This copying approach provides data redundancy and protection against power failures without requiring the original write to be completed to nonvolatile storage before responding to the client, thus improving write response performance while maintaining data permanence through the cached copies.
3Reliability
If data is replicated to multiple nodes to prevent data loss during power interruption, then data safety is improved, but system complexity increases due to coordination requirements
Solution Approach 1:
The patent applies local quality by selecting nodes with different power supply units as transfer destinations. This approach focuses data replication on specific nodes with distinct power sources rather than uniformly replicating to all nodes, providing targeted data safety against power interruptions while reducing system complexity by limiting the replication scope to only those nodes with different power supplies.
Data Source
AI summary
A distributed storage system, which receives a write request from a client, includes a plurality of computers which receive power supply from a plurality of power supply units. A first computer, among the plurality of computers, which is a computer that receives the write request from the client, is configured to: cache updated data which is at least apart of data accompanying the write request; select n second computers which are n computers (n is a natural number) among computers each receiving power from a power supply unit different from a power supply unit of the first computer as transfer destinations of the updated data; and transfer the updated data to the selected n second computers, respectively. At least one of the n second computers, when caching the updated data from the first computer, is configured to return a result to the first computer.