CRUSH Hashing for Dynamic Storage Node Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional storage systems have limited capabilities for dynamically managing storage elements and responding to system failures, particularly in terms of adding or deleting storage components and maintaining data integrity across nodes in a clustered environment.
Innovation Solution
The Saratoga Speed Block Level Cluster (SSBLC) system employs a CRUSH-based hashing library for mapping Logical Unit Numbers (LUNs) across nodes, combined with PAXOS for state coordination, allowing for dynamic storage management and failure recovery by synchronizing node states and remapping data across the cluster.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional storage systems are used, then system simplicity is maintained, but the capability for dynamically adding or deleting storage elements is limited
Solution Approach 1:
The storage system is divided into independent storage elements (disks, partitions, LUNs) that can be individually managed, added, or deleted. Each storage element is treated as a separate unit that can be dynamically provisioned and distributed across multiple nodes in the cluster, enabling flexible storage management without requiring system-wide reconfiguration.
Solution Approach 2:
The system implements dynamic storage management where storage elements can be added or removed at runtime without taking down the entire system. The CRUSH-based hashing library dynamically remaps data chunks when storage topology changes, and the PAXOS protocol dynamically coordinates state transitions across nodes, allowing the system to adapt to changing storage requirements.
2Reliability
If conventional storage systems are used, then system architecture is simple, but the response capability to system failures is limited
Solution Approach 1:
The system pre-establishes redundant copies of data chunks across multiple nodes using CRUSH hashing before failures occur. When a node fails, the system has already positioned replacement copies on other nodes, enabling immediate failover without data loss. The PAXOS protocol pre-coordinates state transitions to ensure consistent failure response across the cluster.
Solution Approach 2:
The system implements continuous monitoring of node health and storage state through PAXOS state coordination. When failures are detected, the system automatically triggers remapping operations and state transitions to restore data availability. This feedback loop enables the system to respond to failures autonomously and maintain reliability without manual intervention.
3Reliability
If data is redistributed across multiple nodes, then data integrity during failures is improved, but data movement overhead increases
Solution Approach 1:
The system replaces traditional mechanical storage allocation with a software-based CRUSH hashing algorithm that deterministically maps data chunks to nodes. This substitution eliminates the need for manual or centralized allocation decisions, enabling automatic and efficient data remapping when failures occur. The hashing-based approach computes new locations instantly without iterative searching or coordination overhead.
Solution Approach 2:
The system changes the storage topology parameters (node additions, removals, or failures) and the CRUSH hashing algorithm automatically recalculates data chunk locations based on the new parameters. This parameter-driven approach allows the system to adapt to changing conditions without manual reconfiguration, minimizing data movement by only relocating chunks affected by the parameter change rather than redistributing entire storage volumes.
Data Source
AI summary
A method of storing a file in a storage system that includes a plurality of memory-storage hosts includes: providing unique chunk identifiers for memory chunks included in the file; using a hash mapping to identify one or more storage locations for each chunk identifier, each storage location corresponding to a portion of a memory-storage host; and storing each memory chuck at the corresponding one or more storage locations identified by the hash mapping.