Virtual Interface Failover Configuration in Clustered Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing virtual interface failover methods in clustered storage systems are limited by the need for multiple read-write transactions, which serialize failover operations and limit concurrency, leading to potential communication bottlenecks and unnecessary interruptions in file system services.
Innovation Solution
Implementing a cluster-wide published configuration system that allows each VIF manager instance to read updated configuration information from a replication database upon failover events, eliminating the need for a separate event table and enabling unaffected managers to confirm their configurations, thus reducing communication bottlenecks and enhancing system robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple read-write transactions are used for VIF failover, then configuration updates can be made, but failover operations are serialized and concurrency is limited
Solution Approach 1:
The patent segments the failover process into two independent parts: (1) a single RW transaction to update the configuration table with new VIF hosting assignments, and (2) multiple independent RO transactions for each VIF manager to read and apply their assigned configurations. This segmentation allows configuration updates to be made once while enabling all VIF managers to simultaneously read and apply their assignments, thereby resolving the contradiction between configuration consistency and failover concurrency.
Solution Approach 2:
The patent extracts the configuration reading operation from the configuration writing operation. Instead of requiring each VIF manager to participate in a coordinated RW transaction, the system extracts the reading phase into independent RO transactions that can execute concurrently. This extraction eliminates the serialization bottleneck while maintaining consistency through the quorum-based RW transaction that establishes the configuration once.
2Ease of operation
If a separate event table is used to notify affected VIF managers, then notification can be targeted, but communication overhead and system complexity increase
Solution Approach 1:
The patent merges the event notification mechanism into the existing configuration table structure. Instead of maintaining a separate event table with additional notification logic, the system combines configuration updates and event signaling into a single configuration table. When the coordinator updates the configuration table with new VIF assignments, this single RW transaction simultaneously serves as both the configuration update and the event notification to all VIF managers, eliminating the need for separate notification infrastructure.
Solution Approach 2:
The configuration table is given multiple functions: it serves as both the storage for VIF hosting assignments and as the event notification mechanism. Every VIF manager monitors this single table for changes, and when updates occur, all managers can determine whether they are affected by checking their local assignments against the updated configuration. This multi-functional approach eliminates the need for separate notification systems while maintaining efficient communication.
3Measurement precision
If RW transactions are used for each VIF manager to read configuration, then configuration accuracy is confirmed, but communication bottlenecks occur
Solution Approach 1:
The patent segments the transaction type based on the operation being performed: a single RW transaction for configuration updates by the coordinator, and multiple concurrent RO transactions for configuration reading by VIF managers. This segmentation allows the system to maintain configuration accuracy through the coordinator's RW transaction while enabling high throughput for the read operations, as RO transactions do not contend for the same database resources.
Solution Approach 2:
The patent changes the transaction parameter from RW (read-write) to RO (read-only) for the configuration reading operations performed by VIF managers. This parameter change fundamentally alters the resource requirements: RO transactions can be performed concurrently without acquiring write locks or contending for database writer resources. The configuration verification accuracy is maintained because VIF managers read from the authoritative configuration table established by the coordinator's RW transaction, while the RO parameter enables high throughput.
Data Source
AI summary
A cluster-wide published configuration system and method is provided for assigning host nodes to virtual interfaces (VIFs) in a cluster of data storage systems. A coordinating VIF manager publishes a set of failover rules in a VIF configuration table using a replication database. In the event of a VIF failure, the coordinating VIF manager publishes a reassignment of host nodes using a single read-write transaction of a replication database service. Secondary VIF managers in the cluster must perform VIF hosting in accordance to the reassignment or publish a new configuration table. Each VIF manager instance in a quorum reads the VIF configuration table upon each VIF failover to perform error checking event regardless of whether the VIF manager is affected by the VIF failover event.