Storage Node Volatile Memory Failover Acceleration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data storage systems face delays in failover procedures due to the need to copy block objects from non-volatile storage to volatile memory and change the state of data from read-only to read-write, which can take several minutes, impacting the speed and efficiency of failover processes in data replication.
Innovation Solution
Implementing a storage node architecture that maintains a replica of a logical volume with its filesystem and metadata in volatile memory and data in non-volatile storage, allowing the state to be changed to read-write in-memory responsive to a failover trigger, and using software-defined network attached storage (SDNAS) applications to synchronize replicas, thereby reducing delays and enabling faster failover.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If block objects are copied from non-volatile storage to volatile memory during failover, then data accessibility is improved, but failover time increases
Solution Approach 1:
The patent maintains filesystem metadata and virtual data mover (VDM) structures in volatile memory of the secondary storage node before failover occurs. This preliminary preparation ensures that when failover is triggered, the secondary node can immediately assume the primary role without requiring time-consuming copying of block objects or reconstruction of filesystem structures, thus resolving the contradiction between data accessibility and failover time.
2Reliability
If data state is changed from read-only to read-write during failover, then data usability is improved, but state transition time increases
Solution Approach 1:
The patent keeps the replica data in a read-write state in the volatile memory of the secondary storage node before failover, rather than in read-only state in non-volatile storage. This preliminary state preparation eliminates the need for time-consuming state transitions during failover, allowing the secondary node to immediately become the active data source upon failover trigger.
3Reliability
If filesystem metadata is maintained in non-volatile storage, then data persistence is improved, but failover speed deteriorates
Solution Approach 1:
The patent maintains filesystem metadata and VDM structures in the volatile memory of the secondary storage node before failover occurs. This preliminary action ensures that all necessary filesystem structures are already loaded and ready in high-speed memory, eliminating the need to read and process metadata from non-volatile storage during failover, thus dramatically improving failover speed while data persistence is maintained through the replication mechanism.
4Reliability
If replica data is kept in read-only state in non-volatile storage, then data integrity is improved, but access speed deteriorates
Solution Approach 1:
The patent maintains an active replica with filesystem metadata and VDM structures in the volatile memory of the secondary storage node before failover. This preliminary preparation ensures that when failover occurs, the replica is immediately accessible at high speed from volatile memory rather than requiring retrieval from non-volatile storage, thus resolving the contradiction between data integrity and access speed.
Data Source
AI summary
A storage node that maintains a replica of a logical volume for use in response to a failover trigger includes a data node with volatile memory in which a filesystem and its metadata and a VDM and its metadata associated with the replica are maintained prior to the failover trigger. The storage node also includes a SAN node in which data associated with the replica is maintained. The data is maintained in a RW (read-write) state by the SAN node prior to the failover trigger. However, the replica is presented in a RO (read-only) state by the storage node prior to the failover trigger. The storage node changes the in-memory state of the filesystem and VDM to RW responsive to the failover trigger. Because the filesystem and its metadata and VDM and its metadata are already in memory and the data is in a RW state in block storage the failover is completed relatively quickly.