CN-121979702-A - Solid state disk residual life prediction method, memory and computer program product

CN121979702ACN 121979702 ACN121979702 ACN 121979702ACN-121979702-A

Abstract

The application discloses a solid state disk residual life prediction method, a memory and a computer program product, and relates to the technical field of storage, wherein the solid state disk residual life prediction method comprises the steps of obtaining a first multidimensional health index related to the service life of a solid state disk, wherein the first multidimensional health index comprises a plurality of health indexes with different dimensions; the method comprises the steps of carrying out index fusion processing on a first multidimensional health index by utilizing a pre-trained convolution self-encoder network to obtain a first fusion health index, determining a fault threshold value corresponding to the first fusion health index by utilizing a preset degradation model, and predicting the residual life of a solid state disk according to the first fusion health index and the fault threshold value by utilizing a pre-trained gating attention unit network. The embodiment of the application can improve the accuracy of predicting the residual life of the solid state disk.

Inventors

SUN CHENGSI
HE HAN
WANG CAN
ZHANG CHAO

Assignees

成都佰维存储科技有限公司

Dates

Publication Date: 20260505
Application Date: 20251203

Claims (14)

1. The solid state disk residual life prediction method is characterized by comprising the following steps of: acquiring a first multidimensional health index associated with the service life of the solid state disk, wherein the first multidimensional health index is a first multidimensional health index The health index comprises health indexes of a plurality of different dimensions; Performing index fusion processing on the first multidimensional health index by using a pre-trained convolution self-encoder network to obtain a first fusion health index; determining a fault threshold corresponding to the first fusion health index by using a preset degradation model; And predicting the residual life of the solid state disk according to the first fusion health index and the fault threshold by using the pre-trained gating attention unit network.
2. The method for predicting remaining life of a solid state disk as claimed in claim 1, wherein the step of determining the failure threshold corresponding to the first fused health indicator using a preset degradation model comprises: Selecting at least one degradation model from a plurality of preset degradation models according to the first fusion health index; Responding to a degradation model, and inputting the first fusion health index into the degradation model to be selected for prediction to obtain a fault threshold; and in response to selecting a plurality of degradation models, determining information criterion information corresponding to the selected degradation models, taking a degradation model corresponding to the smallest information criterion information in the selected degradation models as a target degradation model, and inputting the first fusion health index into the target degradation model to predict to obtain a fault threshold.
3. The method for predicting the remaining life of a solid state disk as claimed in claim 2, wherein the step of selecting at least one degradation model from a plurality of preset degradation models according to the first fused health index comprises: Determining a trend graph comprising a first fused health indicator from a first time point to a second time point, wherein the first time point is less than the second time point; And selecting at least one degradation model from a plurality of preset degradation models according to first trend information corresponding to the first fusion health index represented by the trend graph.
4. The method for predicting remaining life of a solid state disk of claim 3 wherein the degradation model comprises a linear distribution model, an exponential distribution model, and a power law distribution model, The step of selecting at least one degradation model from a plurality of preset degradation models according to the first trend information corresponding to the first fusion health index represented by the trend graph includes at least one of the following steps: Responding to the first trend information, and selecting a linear distribution model from the linear distribution model, an exponential distribution model and a power law distribution model for representing second trend information of the first fusion health index showing stable linear decline along with time change; responding to the first trend information, and selecting an exponential distribution model from the linear distribution model, the exponential distribution model and a power law distribution model for representing third trend information of late-stage accelerated degradation of the first fusion health index; responding to the first trend information, and selecting a power law distribution model from the linear distribution model, the exponential distribution model and the power law distribution model for representing fourth trend information of nonlinear degradation trend caused by sudden load; at least two degradation models are selected among the linear distribution model, the exponential distribution model, and the power law distribution model in response to the first trend information including at least two of the first trend information, the second trend information, and the third trend information.
5. The method of claim 4, wherein the selecting at least two degradation models among the linear distribution model, the exponential distribution model, and the power law distribution model in response to the first trend information including at least two of first trend information, second trend information, and third trend information comprises: determining a demarcation point between different trend information in the trend graph in response to the first trend information including at least two of first trend information, second trend information, and third trend information; Responding to the demarcation point comprising a first demarcation point, for a first fusion health index belonging to the range from the first time point to the first demarcation point, executing the step of selecting at least one degradation model from a plurality of preset degradation models according to the first fusion health index, and for a first fusion health index belonging to the range from the first demarcation point to a second time point, executing the step of selecting at least one degradation model from the plurality of preset degradation models according to the first fusion health index; And executing the step of selecting at least one degradation model from a plurality of preset degradation models according to the first fusion health index, executing the step of selecting at least one degradation model from the plurality of preset degradation models according to the first fusion health index, for the first fusion health index from the second demarcation point to the third demarcation point, and executing the step of selecting at least one degradation model from the plurality of preset degradation models according to the first fusion health index, for the first fusion health index from the third demarcation point to the second demarcation point.
6. The method for predicting the remaining life of a solid state disk according to claim 1, wherein the predicting the remaining life of the solid state disk according to the first fused health indicator and the failure threshold using the pre-trained gated attention unit network comprises: constructing a first hanker matrix comprising a first fused health index of a first time range; Inputting the first Hank matrix into a pre-trained gating attention unit network for prediction, and outputting to obtain a first prediction result; Updating the first prediction result as a first fusion health index of a next time step to the first hanker matrix in response to the first prediction result being less than the fault threshold; According to the updated first Hank matrix, the step of inputting the first Hank matrix into a pre-trained gated attention unit network for prediction is executed until a latest obtained first prediction result is detected to be greater than or equal to the fault threshold; Determining the prediction times of the pre-trained gated attention unit network for prediction according to the latest obtained first prediction result, and determining sampling time corresponding to the first fusion health index, wherein the sampling time comprises sampling duration time and sampling interval time for representing and sampling the first multidimensional health index; and updating the sampling time according to the prediction times to obtain the residual life of the solid state disk.
7. The method for predicting the remaining life of a solid state disk as claimed in claim 1, further comprising at least one of: Model training is carried out on a preset convolutional self-encoder network by using a preset first training data set until a first training cut-off condition is reached, and a pre-trained convolutional self-encoder network is obtained, wherein training data in the first training data set are a second multidimensional health index and a second fusion health index corresponding to the second multidimensional health index; and performing model training on the preset gating attention unit network by using a preset second training data set until a second training cut-off condition is reached, so as to obtain the pre-trained gating attention unit network, wherein training data in the second training data set comprises at least one third fusion health index.
8. The method for predicting remaining life of a solid state disk of claim 7 wherein the convolutional self-encoder network comprises a convolutional encoder and a deconvolution decoder, The step of performing model training on the preset convolutional self-encoder network by using the preset first training data set until a first training cut-off condition is reached, and obtaining a pre-trained convolutional self-encoder network comprises the following steps: Inputting the second multidimensional health index into a preset convolution self-encoder network, and extracting data features of the second multidimensional health index according to the convolution encoder to obtain coding features; reconstructing the coding feature according to the deconvolution decoder to obtain a fourth fusion health index; And calculating a loss function value of the fourth fusion health index and the second fusion health index according to a preset first loss function to obtain a first loss function value, and reversely updating the convolution self-encoder network according to the first loss function value until a first training cut-off condition is reached to obtain a pre-trained convolution self-encoder network.
9. The method for predicting remaining life of a solid state disk of claim 8 wherein the convolutional encoder comprises a convolutional layer, a first activation function, a Dropout layer, a pooling layer, and a fully-connected layer, And extracting data characteristics of the second multidimensional health index according to the convolutional encoder to obtain coding characteristics, wherein the step of extracting the data characteristics of the second multidimensional health index comprises the following steps: respectively carrying out convolution processing on each one-dimensional health index in the second multidimensional health indexes according to the convolution layer to obtain each first convolution result; Respectively performing activation processing on each first convolution result according to the first activation function, performing random inactivation processing on each first convolution result after the activation processing according to the Dropout layer, and performing downsampling on each first convolution result after the random inactivation processing according to the pooling layer to obtain downsampled first convolution results; and performing full connection on each downsampled first convolution result according to the full connection layer to obtain coding features.
10. The method for predicting remaining life of a solid state disk of claim 9, wherein the deconvolution decoder comprises a deconvolution layer, a second activation function and an upsampling layer, The step of reconstructing the coding feature according to the deconvolution decoder to obtain a fourth fusion health index comprises the following steps: Upsampling the coding feature according to the upsampling layer to obtain sampled data; And performing deconvolution activation processing on the sampling data according to the deconvolution layer and the second activation function to obtain a fourth fusion health index.
11. The method for predicting remaining life of a solid state disk of claim 7 wherein the gated attention cell network comprises a reset gate, an update gate, and an attention gate, The step of performing model training on the preset gated attention unit network by using the preset second training data set until reaching a second training cut-off condition to obtain a pre-trained gated attention unit network comprises the following steps: Constructing a second Hanker matrix comprising at least one third fusion health index, and inputting historical information of the second Hanker matrix and the solid state disk to a preset gating attention unit network, wherein the historical information comprises a previous hidden state corresponding to the third fusion health index; Determining reset gate output information according to the second hank matrix and the history information by using the reset gate; Determining update gate output information according to the second hank matrix and the history information by using the update gate; Determining a candidate hidden state according to the reset gate output information, the update gate output information and the history information; Determining an attention distribution matrix according to the reset gate output information and the update gate output information by using the attention gate; Determining a sixth fusion health index according to the candidate hidden state, the reset gate information, the update gate information and the attention distribution matrix; And reversely updating the gating attention unit network according to a preset second loss function and the sixth fusion health index until a second training cut-off condition is reached, so as to obtain a pre-trained gating attention unit network.
12. The method of claim 1-11, wherein the health indicator comprises at least one of a total amount of write data, a daily drive write amount, a write amplification factor, a Nand write amount, a Host write amount, a P/E number, a percentage of used life indicator, a bad block number, a wear level indicator, an available remaining space, an uncorrectable error count, a write error count, an erasure error count, a correctable error count, a cyclic redundancy check error count, a sequential read-write bandwidth performance, a random read-write IOPS performance, a random read-write latency performance, and a random read-write QoS performance.
13. A memory, characterized in that the memory comprises a main control chip and a memory chip, the memory chip stores a computer program, and the main control chip can execute the computer program to implement the solid state disk residual life prediction method according to any one of claims 1 to 12.
14. A computer program product, characterized in that the computer program product comprises a computer program which, when being executed by a processor, implements the steps of the solid state drive remaining life prediction method as claimed in any one of claims 1-12.

Description

Solid state disk residual life prediction method, memory and computer program product Technical Field The present application relates to the field of storage technologies, and in particular, to a method for predicting remaining life of a solid state hard disk, a memory, and a computer program product. Background Solid state disk (Solid STATE DRIVE, SSD) has become a core storage component of a high-performance computing cluster, a distributed storage system (such as OST (Object Storage Target, object storage server)/MDT (METADATA TARGET, metadata server) node of a Lustre parallel file system) and a cloud computing data center due to its low-latency and high-throughput characteristics. However, SSDs are not checked and replaced in time when the service life is exhausted, and especially for enterprise-level customers with larger storage requirements, data can be irreversibly lost, storage system stability is reduced, service downtime and other serious effects are caused, so that huge property and security losses are caused. Monitoring the degradation condition to build Health Indicators (HIs) and make predictions of their remaining useful life (REMAINING USEFUL LIFE, RUL) is therefore critical to ensuring the safety and reliability of SSDs. However, the current method for predicting the remaining service life of the SSD is performed in a manner that depends on a single dimension index (such as the number of erasing times) in the log data, which results in inaccurate predicted remaining service life. The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present application and is not intended to represent an admission that the foregoing is prior art. Disclosure of Invention The application mainly aims to provide a solid state disk residual life prediction method, a memory and a computer program product, which aim to solve the technical problem of how to improve the accuracy of solid state disk residual life prediction. In order to achieve the above objective, the present application provides a method for predicting the remaining life of a solid state disk, the method for predicting the remaining life of the solid state disk comprising: acquiring a first multidimensional health index associated with the service life of the solid state disk, wherein the first multidimensional health index is a first multidimensional health index The health index comprises health indexes of a plurality of different dimensions; performing index fusion processing on the first multidimensional health index by using a pre-trained convolution self-encoder network to obtain a first fusion health index; determining a fault threshold corresponding to the first fusion health index by using a preset degradation model; And predicting the residual life of the solid state disk according to the first fusion health index and the fault threshold by using the pre-trained gating attention unit network. In an embodiment, the step of determining the fault threshold corresponding to the first fused health indicator using a preset degradation model includes: selecting at least one degradation model from a plurality of preset degradation models according to the first fusion health index; Responding to the selection of one degradation model, and inputting a first fusion health index into the selected degradation model to predict to obtain a fault threshold; and in response to the selection of the multiple degradation models, determining information criterion information corresponding to the selected multiple degradation models, taking a degradation model corresponding to the minimum information criterion information in the selected multiple degradation models as a target degradation model, and inputting a first fusion health index into the target degradation model for prediction to obtain a fault threshold. In an embodiment, the step of selecting at least one degradation model from a plurality of preset degradation models according to the first fused health index comprises: Determining a trend graph comprising a first fused health indicator from a first time point to a second time point, wherein the first time point is less than the second time point; and selecting at least one degradation model from a plurality of preset degradation models according to first trend information corresponding to the first fusion health index represented by the trend graph. In one embodiment, the degradation model includes a linear distribution model, an exponential distribution model, and a power law distribution model, According to first trend information corresponding to a first fusion health index represented by a trend graph, selecting at least one degradation model from a plurality of preset degradation models, wherein the step comprises at least one of the following steps: Responding to the first trend information, and selecting a linear distribution model from the linear distribution model, the exponential