Storage Saturation Detection and Admission Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional IOPS benchmarks for storage systems are inadequate in characterizing performance under dynamic conditions and diverse workloads, leading to over-provisioned data center storage systems and increased costs due to poor prediction of performance for online systems operating outside narrow ranges.
Innovation Solution
A method to estimate maximum throughput and workload latency by monitoring storage unit performance over time, determining a threshold latency for workload admission, and implementing load balancing and quality of service policies to optimize storage system utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional IOPS benchmarks are used to characterize storage systems, then performance estimates can be obtained for narrow operating ranges, but the characterization is imprecise and cannot predict performance for diverse workloads or dynamic conditions
Solution Approach 1:
The patent transitions from static benchmarking to dynamic performance characterization by continuously monitoring storage systems under varying workload conditions. The system adapts to different workload types and intensities in real-time, capturing performance metrics across the full operating range rather than at fixed benchmark points.
Solution Approach 2:
The patent changes the fundamental parameters used for characterization from fixed IOPS benchmarks to a combination of latency and throughput measurements taken under diverse workload conditions. This allows the system to model performance across varying load levels, mixing ratios, and access patterns.
2Reliability
If worst-case performance estimates are used to guarantee required overall performance, then performance requirements can be met under all conditions, but storage systems become extremely over-provisioned and costs increase dramatically
Solution Approach 1:
The patent implements continuous performance monitoring and modeling that provides feedback on actual system behavior under varying conditions. This enables dynamic capacity planning and admission control decisions based on real-time performance characteristics rather than static worst-case assumptions, optimizing the balance between reliability and resource utilization.
Solution Approach 2:
The patent performs preliminary performance characterization and modeling during system operation to establish accurate performance baselines before making capacity planning or admission control decisions. This preliminary data collection enables more accurate predictions of future performance under various provisioning scenarios.
3Ease of manufacture
If static benchmarks are used for performance characterization, then implementation is simple, but the benchmarks cannot reflect dynamic relationships that impact online system performance
Solution Approach 1:
The patent enables the storage system to characterize its own performance through automated monitoring and analysis of its operational data. The system collects performance metrics during normal operation, builds performance models, and uses these models for capacity planning and optimization without requiring external benchmarking interventions.
Data Source
AI summary
Maximum throughput of a storage unit, and workload and latency values of the storage unit corresponding to a predefined fraction of the maximum throughput are estimated based on workloads and latencies that are monitored on the storage unit. The computed metrics are usable in a variety of different applications including admission control, storage load balancing, and enforcing quality of service in a shared storage environment.