Distributed Statistics Management in Scale-Out Storage Clusters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In distributed storage systems, the principal node can become a bottleneck as the cluster scales up, impacting stability and efficiency in tracking statistics across the cluster.
Innovation Solution
Implementing a flat architecture that distributes cluster-level functions evenly among nodes, allowing for the initialization of statistics groups on multiple storage nodes, assigning ownership, collecting statistics, and combining values using functions like sum, latest, min, or max, while employing a microservice architecture with a statistics manager and clients to manage statistics efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a dedicated principal node is used to coordinate cluster activity and track statistics, then cluster-level functions are centralized and easy to manage, but the principal node becomes a bottleneck as the cluster scales up, impacting stability and performance
Solution Approach 1:
The patent segments the centralized statistics management function by introducing virtual statistics groups that can be distributed across multiple storage nodes. Each statistics group is associated with a managed object and can be owned by different nodes, allowing statistics collection to be divided and distributed rather than concentrated at a single principal node.
Solution Approach 2:
The patent introduces an intermediary mechanism where storage nodes act as both data collectors and statistics managers for their respective managed objects. The virtual statistics groups serve as intermediaries that aggregate statistics from multiple nodes without requiring a dedicated principal node, enabling distributed statistics management while maintaining coordination.
2Reliability
If a flat architecture is used to distribute cluster-level functions among nodes, then the system can scale to thousands of nodes without a bottleneck, but tracking statistics across distributed nodes becomes more complex
Solution Approach 1:
The patent implements universal statistics groups that can serve multiple purposes: they can be owned by individual nodes for local statistics collection, shared across nodes for aggregated statistics, and queried by any node in the cluster. This multi-functionality simplifies the distributed statistics tracking mechanism by providing a unified interface regardless of the underlying distribution complexity.
Solution Approach 2:
The patent implements feedback mechanisms where storage nodes continuously report statistics to the virtual statistics groups, and the system provides feedback through standardized APIs that allow any node to query aggregated statistics. This feedback loop maintains consistency and simplifies tracking by automatically aggregating and distributing statistics information throughout the cluster.
3Productivity
If statistics are collected on multiple storage nodes in a flat architecture, then the system achieves massive scale-out capability, but combining statistics values from multiple nodes requires additional processing overhead
Solution Approach 1:
The patent applies preliminary action by pre-computing and caching statistics values at each storage node before they need to be aggregated. Statistics groups maintain local copies of relevant statistics data, and aggregation operations retrieve and combine these pre-computed values rather than collecting raw data from all nodes, significantly reducing the processing overhead for combining statistics.
Data Source
AI summary
Systems and processes for statistics management in a distributed storage system using a flat cluster architecture. Statistics for managed objects are collected using virtual statistics groups across multiple storage nodes. The systems and processes are compatible with storage systems that utilize microservice architectures.