Storage Device Degradation Detection via I/O Completion Time Distribution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting hard drive degradation in high-performance computing systems are intrusive, resource-intensive, and can reduce performance, making it difficult to predict and detect reduced performance in a non-intrusive manner.
Innovation Solution
A method that monitors storage device parameters such as throughput, queue time, and completion time, generating a model of expected distributions during normal operation and comparing them to actual distributions to identify faults, using Bayesian probability and Poisson distributions to analyze correlations and identify deviations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If SMART diagnostics are performed on hard drives to detect degradation and failures, then detection capability is improved, but system performance and read/write speed are reduced
Solution Approach 1:
The patent introduces an intermediary statistical model that indirectly monitors drive health by analyzing the distribution patterns of I/O completion times. Instead of directly querying drive status through SMART diagnostics (which impact performance), the system uses completion time distributions as a mediator to infer drive degradation, thereby avoiding the performance penalty of direct diagnostic commands.
Solution Approach 2:
The patent replaces the mechanical SMART diagnostic process with a statistical analysis approach. Instead of executing diagnostic commands that mechanically interact with the drive, the system substitutes this with computational analysis of naturally occurring I/O completion time data, eliminating the need for intrusive mechanical diagnostics while maintaining detection capability.
2Measurement precision
If SMART data collection is performed frequently to monitor drive status, then detection sensitivity is improved, but system load and resource consumption increase
Solution Approach 1:
The system uses self-service by leveraging existing I/O completion time data that is naturally generated during normal drive operation. Instead of requiring additional diagnostic commands or resources to monitor drive health, the system repurposes the completion time information already being generated as part of normal operations, making the monitoring process self-sustaining without additional resource burden.
3Reliability
If traditional SMART monitoring is used to detect drive failures, then fault detection is achieved, but subtle performance degradations are missed
Solution Approach 1:
The patent applies preliminary action by detecting subtle distribution changes in completion times before they manifest as outright failures. By continuously analyzing the statistical distribution of completion times, the system identifies early signs of degradation in the distribution pattern, allowing for preventive action before the drive completely fails, thus improving both reliability and detection precision.
Data Source
AI summary
A method of monitoring the performance of a device is presented, each storage device being operable to receive storage requests and includes a queue for incoming storage requests and a physical storage medium, the method includes: a) obtaining data relating to the device parameters of: throughput of storage requests to the storage device; queue time for one or more requests on the storage device; and the time required to complete one or more storage requests on the storage device; b) generating a model relating to the expected distribution of the device parameters on the storage device during normal operation; c) monitoring the parameters during operation of the storage device; d) comparing the distribution of the monitored parameters during operation of the storage device with the expected distribution of the parameters; and e) identifying a fault on the storage device based on the comparison.