Dynamic Disk Load Balancing via Spare Throughput
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional load balancing techniques for scalable storage systems fail to efficiently distribute workloads across disks, often leading to inefficiencies due to static mappings, over-correction, and neglect of factors like disk size and throughput variations.
Innovation Solution
A dynamic load balancing method that calculates spare throughput on each disk, transferring workloads from heavily loaded disks to lightly loaded ones only if the lightly loaded disks have higher than average spare throughput and the average spare throughput is not decreasing, ensuring efficient workload distribution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Stability of the object's composition
If traditional static load mapping is used to keep each client with the same disk, then client-disk association is simple and stable, but workload distribution becomes inefficient due to slow clients and overhead
Solution Approach 1:
The patent implements dynamic load balancing that continuously monitors workload metrics (spare throughput, read/write percentages) and adjusts client-disk mappings in real-time. This allows the system to adapt to changing client performance characteristics and workload patterns, resolving the contradiction between stability and efficiency by making the mapping dynamic rather than static.
Solution Approach 2:
The system employs feedback mechanisms by calculating spare throughput and workload distribution metrics, then using this information to make informed decisions about workload transfers. The feedback loop ensures that load balancing actions are based on actual system state, improving efficiency while maintaining stability through data-driven adjustments.
2Productivity
If traditional dynamic load balancing transfers workloads to balance loads, then workload distribution improves, but the system over-corrects causing perpetual oscillations between heavily and lightly loaded disks
Solution Approach 1:
The patent applies preliminary anti-action by establishing threshold-based transfer conditions that prevent unnecessary workload transfers. The system only initiates transfers when specific criteria are met (spare throughput differences exceed thresholds, average spare throughput is not decreasing), which prevents over-correction and oscillations while still achieving effective load balancing when needed.
Solution Approach 2:
The system changes the parameter used for load balancing from simple load metrics to spare throughput calculations. By using spare throughput as the primary metric and incorporating multiple transfer conditions, the system achieves more stable workload distribution that avoids the oscillation problems of traditional dynamic load balancing.
3Device complexity
If traditional dynamic load balancing is used without considering disk size, then implementation is simple, but workload distribution becomes inefficient across differently sized disks
Solution Approach 1:
The patent applies local quality by tailoring the load balancing approach to each disk's specific characteristics, particularly disk size. The system calculates spare throughput considering individual disk capacities and transfers workloads proportionally, ensuring that larger disks can accept more workload while smaller disks maintain appropriate headroom. This resolves the contradiction by making the load balancing adaptive to local disk properties.
Data Source
AI summary
The disclosed computer-implemented method for dynamic load balancing on disks may include (1) calculating the spare throughput for each disk, (2) identifying a lightly loaded disk and a heavily loaded disk, (3) identifying a set of workloads to be transferred from the heavily loaded disk to the lightly loaded disk by: (a) beginning with the set empty, (b) identifying candidate workloads on the heavily loaded disk, (c) adding a new workload from the candidate workloads to the set when the new workload would not reduce the spare throughput on the lightly loaded disk below a threshold if both the set and the workload were transferred to the lightly loaded disk, and (d) considering each workload for transfer in order from most throughput consumed to least throughput consumed, and (4) transferring the set of workloads. Various other methods, systems, and computer-readable media are also disclosed.