Abstract: Techniques for determining vulnerability of disks are described herein. According to one embodiment, for each of a plurality of disks representing a redundant array of independent disks (RAID), a reallocated sector count associated with the disk is obtained, the reallocated sector count representing a number of sectors that have been reallocated due to an error of a storage transaction to the disk. A failure probability of the disk given the obtained reallocated sector count is determined using a predictive model, wherein the predictive model was generated based on history operating data of a set of known disks. Thereafter, a failure probability of at least two of the disks in the RAID is determined based on the failure probability of each of the disks to determine vulnerability of the RAID.

Abstract: According to one embodiment, a diagnostic parameter of a target storage disk of a storage system is received. A first weight factor is determined from a first quantile distribution representation based on a value of the diagnostic parameter, where the first quantile distribution representation represents a quantile distribution of values of the diagnostic parameter of a set of known failed disks. A second weight factor is determined from a second quantile distribution representation based on the value of the diagnostic parameter, where the second quantile distribution representation represents a quantile distribution of values of the diagnostic parameter of a set of known working disks. A third weight factor is determined for the diagnostic parameter for the target storage disk based on the first weight factor and the second weight factor and a probability of potential disk failure of the target storage disk is determined based on the third weight factor.

Abstract: Techniques for determining a disk failure indicator for predicting disk failures are described herein. According to one embodiment, diagnostic parameters are received which are collected from a set of known working disks and a set of known failed disks of a storage system. For each of the diagnostic parameters, a first quantile distribution representation is generated for the set of known working disks, and a second quantile distribution representation is generated for the set of known failed disks. The first quantile distribution representation and the second quantile distribution representation of each of the diagnostic parameters are then compared to select one or more of the diagnostic parameters as one or more disk failure indicators for predicting future disk failures.

Abstract: Techniques for predicting single-disk failures are described herein. According to one embodiment, first values of a predetermined diagnostic parameter collected from a set of known working disks and a set of known failed disks of a storage system are received. An optimal threshold of the received first values of the predetermined diagnostic parameter is determined based on a quantile distribution of the first values against a plurality of threshold candidates. A potential risk of disk failures of one or more disks of a target storage system is determined based on second values of the predetermined diagnostic parameter collected from the one or more disks of the target storage system in view of the optimal threshold.

Abstract: Techniques for predicting multiple disk failures are described herein. According to one embodiment, first values of a predetermined diagnostic parameter collected from a set of known failed disks of a storage system are received. A quantile distribution graph of the first values against percentiles of a number of known failed disk is generated. In response to second values of the predetermined diagnostic parameter collected from a set of target disks, each of the second values is applied to the quantile distribution graph to determine a corresponding percentile for each of the target disks, which represents a failure probability of a corresponding one of the target disks. A failure probability of two or more of the target disks is calculated based on the determined percentiles of the target disks.

Abstract: Techniques for disk scrubbing are described herein. According to one embodiment, for each of the sectors of a storage disk, a read operation is issued to a sector of a first storage disk of a storage system. In response to an error returned from the read operation, it is determined whether the sector is a live sector, first data is generated if the sector is not a live sector, and the first data is written to the sector to trigger a remap process that remaps a logical block address associated with the sector to another physical block address. The remap process modifies the reallocated sector count. It is determined whether the first storage disk is vulnerable based on a number of reallocated sectors resulted from the first storage disk based on the remap process, which is performed by internal logic of the first storage disk.

Abstract: Techniques for disk scrubbing are described herein. According to one embodiment, for each of the sectors of a storage disk, a read operation is issued to a sector of a first storage disk of a storage system. In response to an error returned from the read operation, it is determined whether the sector is a live sector, first data is generated as random data if the sector is not a live sector, and the first data is written to the sector to trigger a remap process that remaps a logical block address associated with the sector to another physical block address. The remap process increments the reallocated sector count. It is determined whether the first storage disk is vulnerable based on a number of reallocated sectors resulted from the first storage disk based on the remap process, which is performed by internal logic of the first storage disk.