Storage Cluster Performance Class Access Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data storage systems face issues with inconsistent client performance due to hot spots where some drives are over-utilized while others are under-utilized, leading to unpredictable performance and difficulty in managing load across the cluster.
Innovation Solution
The system implements a performance management method that determines client metrics and system metrics to calculate a load value, adjusting the performance of clients based on target performance values and quality of service parameters, ensuring even distribution of data across drives to prevent hot spots and maintain consistent aggregate performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data is stored on a small percentage of drives in a storage cluster, then storage capacity is efficiently utilized, but hot spots occur where portions of the cluster are over-utilized while other portions are under-utilized
Solution Approach 1:
The patent segments the storage cluster into multiple performance classes (e.g., premium, standard, economy) with different performance guarantees. Each class has defined IOPS (input/output operations per second) guarantees, allowing the system to divide the storage resource pool into manageable segments with differentiated service levels. This segmentation resolves the contradiction by enabling efficient capacity utilization in economy classes while providing performance consistency guarantees in premium classes.
Solution Approach 2:
The patent applies local quality by assigning different performance characteristics to different portions of the storage cluster. Specifically, it creates performance classes with varying IOPS guarantees and allows administrators to allocate specific drives or storage pools to different performance classes. This enables hot spots to be contained to specific segments while maintaining overall system efficiency, as under-utilized drives can be allocated to premium performance classes while economy classes use standard capacity drives.
2Ease of operation
If quality of service based client prioritization is implemented, then client experience is improved, but specific performance levels are not guaranteed and system performance becomes unpredictable under stress
Solution Approach 1:
The patent implements dynamic performance management through performance classes that can be flexibly configured and adjusted. The system dynamically allocates IOPS guarantees based on the performance class assigned to each client, rather than using static prioritization. This allows the system to maintain predictable performance under stress by enforcing IOPS limits and guarantees at the performance class level, while still providing differentiated client experiences through class selection.
Solution Approach 2:
The patent changes the fundamental parameter from simple client prioritization to performance class assignment with explicit IOPS guarantees. By introducing measurable performance parameters (IOPS, bandwidth, latency) as defining characteristics of each performance class, the system transforms subjective quality of service into objective, guaranteed performance levels. This allows administrators to predictably control system performance under various load conditions.
3Productivity
If prioritization system is used to divide full performance among all clients, then access requests are processed based on priority, but single client effects on cluster performance are not capped and system always runs slow under stress
Solution Approach 1:
The patent applies preliminary anti-action by pre-defining performance class boundaries and IOPS guarantees before clients access the storage system. Each performance class has predetermined performance characteristics that cap the maximum impact any client can have on cluster performance. This prevents any single client from overwhelming the system, as their access is limited by their assigned performance class parameters, while still allowing high-throughput processing for clients in appropriate performance classes.
Data Source
AI summary
Performance of a storage system with data distributed substantially, evenly across a cluster of storage nodes can be dynamically managed according whether quality of service (QoS) rules based on one or more performance capacities according to performance class is satisfied, wherein each QoS rule includes a boundary condition and a threshold expressing performance guarantees to a respective performance class. Determining for each QoS rule, whether a respective boundary condition satisfies a respective threshold based on measured capacity for a performance metric corresponding to the respective boundary condition; and if so, setting a first QoS parameter associated with the set of QoS rules and throttling access to the storage system by a first client associated with a first performance class based on the first QoS parameter to enforce the performance guarantees to the first performance class.