Erasure-Coded Fragment Migration to Eliminate Node Co-Hosting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Erasure coding (EC) systems with fewer than data-plus-parity storage nodes are vulnerable to data loss and service disruptions due to node failures, as they lack sufficient resources for reconstruction and distribution during failures.
Innovation Solution
A distributed data storage system that grows from fewer than data-plus-parity nodes to at least D+P nodes, migrates data fragments to ensure no co-hosting on the same node, and switches between erasure coding and replication modes to maintain data integrity and availability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If the system uses fewer than data-plus-parity storage nodes, then storage resources are conserved and storage efficiency is improved, but the system becomes vulnerable to node failures and data loss
Solution Approach 1:
The system performs preliminary actions by detecting when the number of available storage nodes reaches or exceeds the data-plus-parity count, and proactively initiates migration of storage containers from co-hosted nodes to newly added nodes before failures occur. This preparatory migration ensures that the system transitions from a vulnerable state (fewer than D+P nodes) to a resilient state (at least D+P nodes) in advance, eliminating the vulnerability to concurrent multi-node failures without requiring a system restart.
Solution Approach 2:
The system dynamically adapts its data distribution strategy based on the number of available storage nodes. When nodes are added and the count reaches D+P, the system automatically triggers a migration process that redistributes storage containers across distinct nodes. This dynamic behavior allows the system to optimize between storage efficiency (when having fewer nodes) and reliability (when having sufficient nodes), transitioning between states as nodes are added or removed.
2Reliability
If the system migrates data fragments to ensure no co-hosting on the same node, then data availability and resilience are improved, but system complexity and migration overhead increase
Solution Approach 1:
The migration process is designed to be self-service, operating automatically without requiring system restarts or manual intervention. The system autonomously detects when migration conditions are met (when node count reaches D+P), identifies co-hosted storage containers, and executes the migration to newly added nodes. This self-service approach reduces operational complexity while achieving the goal of distributing data fragments across distinct nodes for improved data availability.
Solution Approach 2:
The system introduces an intermediary migration mechanism that facilitates the transfer of storage containers from co-hosted nodes to newly added nodes. This intermediary process acts as a bridge between the existing data distribution state and the desired distributed state, managing the complexity of data migration while ensuring data availability. The intermediary migration logic handles the redistribution of erasure-coded fragments without requiring system downtime or complex manual procedures.
3Productivity
If the system co-hosts storage containers on the same node, then device utilization is improved, but the system becomes vulnerable to simultaneous failure of multiple storage containers
Solution Approach 1:
The system extracts storage containers from co-hosted nodes and relocates them to newly added nodes, separating previously co-located storage containers onto distinct nodes. This extraction process eliminates the harmful factor of simultaneous failure risk by ensuring that storage containers for the same virtual disk are distributed across different nodes, while the migrated containers continue to utilize available device resources effectively.
Solution Approach 2:
The system transitions the data distribution from a single-dimension co-hosting model to a multi-dimensional distributed model. By migrating storage containers to newly added nodes, the system adds a new dimension to the data distribution architecture, spreading containers across multiple nodes in the cluster. This dimensional change reduces the concentration of storage containers on single nodes, thereby reducing the risk of simultaneous failures while maintaining device utilization.
Data Source
AI summary
A distributed data storage system that employs erasure coding grows from fewer than data-plus-parity (D+P) storage service nodes to at least D+P nodes. The system detects an increase in the number of available storage service nodes, i.e., at least D+P; analyzes how storage for each virtual disk is distributed among storage containers in the existing (pre-growth) nodes; identifies containers that are co-hosted on the same node; and, on a node-by-node basis, migrates data fragments from a co-hosted storage container to a corresponding new container that is configured on another node. For each virtual disk in the illustrative system, the migration causes containers, including the erasure-coded data fragments they host, to be re-distributed so that the containers for a virtual disk are NOT doubled up or co-hosted on the same node. The disclosed computer-implemented process for re-distributing erasure-coded data fragments operates organically, without requiring the system to restart or reboot.