Load Balancing Virtual Data Movers in Storage Clusters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing storage cluster systems face challenges in efficiently balancing the workload across nodes to maintain optimal performance and reliability, particularly in managing virtual data movers (VDMs) due to uneven node scores and resource utilization.
Innovation Solution
A method and apparatus for load balancing VDMs between nodes in a storage cluster, where a cluster manager collects and weights node statistics to assign scores, applies hard and soft rules to identify optimal VDM movement combinations, and implements these movements to equalize node scores, thereby distributing workload evenly across nodes while minimizing resource-intensive VDM reconfiguration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If VDMs are assigned to nodes in a storage cluster, then data movement and storage operations can be performed, but workload becomes unevenly distributed across nodes leading to performance degradation
Solution Approach 1:
The system dynamically reassigns Virtual Data Movers between nodes based on real-time node scores that reflect current workload conditions. The cluster manager continuously evaluates node statistics and adjusts VDM assignments to maintain balanced workload distribution, transforming a static assignment model into a dynamic one that adapts to changing conditions.
Solution Approach 2:
The system implements a feedback mechanism where node statistics are collected, node scores are calculated, and these scores feed back into VDM reassignment decisions. The cluster manager uses the node scores to identify optimal VDM movement combinations, creating a closed-loop control system that continuously optimizes workload distribution based on system state feedback.
2Ease of operation
If VDM reconfiguration is performed to balance workload, then node score equalization is achieved, but resource-intensive operations and system disruption occur
Solution Approach 1:
The system performs partial reconfiguration by moving only the necessary subset of VDMs that will achieve workload balance, rather than reconfiguring all VDMs. The cluster manager identifies specific VDM movement combinations that equalize node scores with minimal movement, applying partial action to reduce resource consumption while achieving the balancing objective.
Solution Approach 2:
The system changes the operational parameters of VDM assignments by adjusting which VDMs are assigned to which nodes based on node scores. The cluster manager modifies assignment parameters selectively to achieve score equalization, changing only the necessary parameters rather than performing comprehensive reconfiguration, thereby reducing resource overhead.
3Measurement precision
If comprehensive node statistics collection is implemented, then accurate node scoring is achieved, but system complexity and overhead increase
Solution Approach 1:
The system uses a universal node statistics collection mechanism that gathers multiple types of information (CPU utilization, memory usage, storage I/O, network traffic) through a single integrated framework. The cluster manager employs a multi-functional approach to collect and process diverse statistics, reducing overall system complexity while maintaining comprehensive monitoring and accurate node scoring.
Data Source
AI summary
Data Virtual Data Movers (VDM) are assigned to nodes of the storage cluster and a backup node is assigned for each data VDM. A system VDM on each node collects node statistics including operational parameters of the node and activity levels of the data VDMs on the node. A cluster manager collects the node statistics from each of the system VDMs and uses weighted collected node statistics to assign a node score to each node in the storage cluster. The cluster manager uses the node scores to identify possible data VDM movement combinations within the storage cluster by applying a set of hard rules and a set of soft rules to evaluate the possible data VDM movement combinations. If a VDM movement combination is selected, it is implemented by moving at least some of the data VDMs within the cluster to attempt to equalize node scores within the cluster.