TWI790938B - Creating method of a classifying model of a efficiency problem of a hard disk, analyzing method of an efficiency problem of a hard disk and classifying model creating system of the efficiency problem of a hard disk - Google Patents

Abstract

A creating method of a classifying model of a efficiency problem of a hard disk includes performing the following steps by a processing device: acquiring a plurality of measured data of a plurality of single hard disk and each of the plurality of measured data comprising values of a plurality of vibration parameters; discretizing the plurality of measured data based on a K-means algorithm; and obtaining the classifying model of the efficiency problem of the hard disk based on the plurality of measured data after discretizing and a decision tree algorithm.

Description

Method for establishing hard disk performance problem classification model, hard disk performance problem analysis method and hard disk performance problem classification model building system

The invention relates to a method for establishing a classification model, in particular to a method for establishing a classification model for hard disk performance problems.

With the development of the Internet, the amount of information processing is increasing, and the number of servers required is also increasing. If the performance of the server is degraded, it will affect the data transmission of the entire network, which may cause the game to crash, the normal transmission of emails or the interruption of the video conference. How to improve the degraded performance of the server has become an important issue.

Generally speaking, the performance of the server is related to the performance of the hard disk, and the performance of the server will be affected by the performance of the hard disk. If the performance of the hard disk decreases, the performance of the server will also decrease. Engineers often use manual step-by-step analysis to find the reasons for hard disk performance degradation inside the server. It often happens that the root cause that affects hard disk performance cannot be found, so that it is impossible to prescribe the right medicine to solve the problem of hard disk performance degradation.

According to the foregoing, the present invention provides a method for establishing a classification model of hard disk performance problems, a method for analyzing hard disk performance problems, and a system for establishing a classification model of hard disk performance problems, so as to find out the root causes affecting the performance of hard disks.

The method for establishing a hard disk performance problem classification model according to an embodiment of the present invention is executed by an analysis device, which includes: obtaining multiple measurement data of multiple single hard disks, each measurement data includes multiple vibrations multiple values of parameters; based on k-means algorithm (k-means algorithm) to discretize these measurement data; based on these discretized measurement data and decision tree algorithm, obtain hard disk performance problem classification model.

A method for analyzing hard disk performance problems according to an embodiment of the present invention includes: using a computer system to input measurement data of abnormal server hard disks into the aforementioned hard disk performance problem classification model to obtain classification results; wherein the classification results An ordering among a plurality of problem factors is indicated, and the problem factors are associated with many of the plurality of vibration parameters.

A hard disk performance problem classification model building system according to an embodiment of the present invention includes a plurality of vibration parameter sensors and an analysis device. A plurality of vibration parameter sensors is used for measuring a plurality of values of a plurality of vibration parameters of each of the plurality of single hard disks. The analysis device is connected to these vibration parameter sensors and is used to obtain multiple measurement data of multiple single hard disks. The analysis device discretizes these measurement data based on the k-mean algorithm, and the analysis device is based on these discretized measurements. The data and the decision tree algorithm are used to obtain a hard disk performance problem classification model, wherein each measurement data includes multiple values of multiple vibration parameters corresponding to multiple single hard disks.

In summary, the method for establishing a classification model of hard disk performance problems and the system for establishing a classification model of hard disk performance problems of the present invention use the k-average algorithm to discretize the vibration measurement data of multiple single hard disks, and combine Decision tree algorithm can establish a decision tree with high classification accuracy as a hard disk performance problem classification model, and since the vibration measurement data of a single hard disk is used as training data, the established decision tree is pruned The demand is not high. In addition, in the hard disk performance problem analysis method of the present invention, the measurement data of the hard disk with problems inside the servo storage system is input into the above-mentioned hard disk performance problem classification model, and the main reason for the hard disk performance degradation can be predicted well.

The detailed features and advantages of the present invention are described in detail below in the implementation mode, and its content is enough to make any person familiar with the related art understand the technical content of the present invention and implement it accordingly, and according to the content disclosed in this specification, the scope of the patent application and the drawings , anyone skilled in the art can easily understand the purpose and advantages of the present invention. The following examples are to further describe the concept of the present invention in detail, but not to limit the scope of the present invention in any way.

It should be understood that although the terms "first", "second" and the like may be used in the present invention to describe various elements, components, regions, layers and/or sections, these elements, components, regions, layers and/or sections Should not be limited by these terms. These terms are only used to distinguish one element, component, region, layer and/or section from another element, component, region, layer and/or section.

In addition, the terms "comprising" and/or "comprising" refer to the presence of stated features, regions, integers, steps, operations, elements and/or parts, but do not exclude one or more other features, regions, integers, steps, operations , the presence or addition of elements, parts and/or combinations thereof.

Please refer to FIG. 1 , which is a functional block diagram of a hard disk performance problem classification model building system according to an embodiment of the present invention. As shown in FIG. 1 , the hard disk performance problem classification model building system 1 includes a plurality of vibration parameter sensors 11 and an analysis device 12 , wherein the analysis device 12 is connected to the vibration parameter sensors 11 .

The single hard disk referred to below refers to the hard disk device that can be loaded on the server. One server can be loaded with multiple hard disk devices. According to its expandability, it can be loaded with 4, 8, or more than 12 hard disks. disc device. The multiple vibration parameter sensors 11 are used to measure multiple values of multiple vibration parameters of each of multiple single hard disks (single disk). Further, the plurality of vibration parameter sensors 11 may include at least two of a three-axis accelerometer 111 , a sound pressure meter 112 and a resonance frequency analyzer 113 . The three-axis accelerometer 111 is used for measuring acceleration values and angular acceleration values of multiple single hard disks. The sound pressure meter 112 is used to measure the sound pressure values of multiple single hard disks. The resonance frequency analyzer 113 can be realized by a hard disk I/O performance evaluation tool (IOMeter) or a spectrum analyzer, and is used to analyze the resonance frequency values of multiple single hard disks. That is, the plurality of vibration parameters may include at least two of acceleration, angular acceleration, sound pressure and resonance frequency.

The analysis device 12 may be a microcontroller, a graphics processor or other electronic devices with functions of processing data and storing functions, and is not limited to the scope of the present invention. The analysis device 12 is used to obtain multiple pieces of measurement data of the multiple single hard disks, wherein each piece of measurement data includes multiple values of multiple vibration parameters of the corresponding ones of the multiple single hard disks, That is, the vibration parameter values measured by the aforementioned multiple vibration parameter sensors 11 . The analysis device 12 discretizes the above multiple measurement data based on the k-means algorithm, and obtains a hard disk performance problem classification model based on the discretized measurement data and a decision tree algorithm.

In an implementation aspect, multiple vibration sensors 11 and analysis devices 12 are integrated into one electronic device. In another embodiment, the plurality of vibration sensors 11 and the analysis device 12 are set independently, and the analysis device 12 can be set at the edge or cloud and communicated with the plurality of vibration sensors 11 .

Please refer to FIGS. 1 and 2 together, which are flowcharts of a method for establishing a hard disk performance problem classification model according to an embodiment of the present invention. As shown in FIG. 2 , the method for establishing a classification model of hard disk performance problems includes steps S1 to S3. The hard disk performance problem classification model building method shown in FIG. 2 is applicable to the hard disk performance problem classification model building system 1 shown in FIG. 1 , but not limited thereto. The following exemplarily uses the operation of the hard disk performance problem classification model building system 1 shown in FIG. 1 to illustrate steps S1 to S3.

Step S1: Using the analysis device 12 to obtain multiple pieces of measurement data of multiple single hard disks, each piece of measurement data includes multiple values of multiple vibration parameters. As mentioned above, the analysis device 12 can obtain the values of multiple vibration parameters measured by the multiple vibration sensors 11 for each single hard disk. In an embodiment, the analysis device 12 can control the plurality of vibration sensors 11 to measure each single hard disk and return the measurement result. In another embodiment, the plurality of vibration sensors 11 can be controlled by the user or other control devices to measure each single hard disk, and then transmit the measurement results to the analyzing device 12 .

Specifically, the three-axis accelerometer 111 can be controlled to detect the acceleration value and angular acceleration value of each single hard disk in the x-axis direction, y-axis direction and z-axis direction, especially the acceleration and angular acceleration in the x-axis direction. The value of the angular acceleration in the z-axis direction, and the detection result is sent to the analysis device 12 . For example, the structure of a single hard disk is a cuboid, the x-axis direction is parallel to the short side direction of the cuboid, the y-axis direction is parallel to the long side direction of the cuboid, and the z-axis is parallel to the height direction of the cuboid, but not This is the limit. The sound pressure meter 112 can be controlled to detect the sound pressure value of each single hard disk, especially the sound pressure value when the single hard disk is fully rotated, and transmit the detection result to the analysis device 12 . The resonant frequency analyzer 113 can be controlled to sweep the frequency of each single hard disk to obtain the resonant frequency value, wherein the sweeping frequency range is, for example, 50 Hz-2000 Hz. Alternatively, the user can set the value of the resonance frequency to 300 Hz or 900 Hz through the user interface connected to the analyzing device 12 .

， 更新所有

，

其中，

為判別量測資料是否屬於第i群體的結果，

為第 i群體的中心，

為第n個量測資料。特別來說，其中i值可以設定為4。 Step S2: using the analysis device 12 to discretize the plurality of measurement data based on the k-mean algorithm. Specifically, the analysis device 12 discretizes the values of different vibration parameters in the measurement data by using a k-mean algorithm to divide them into multiple groups. The k-average algorithm is represented by the following formula, for example:

, to update all

,

in,

In order to judge whether the measurement data belong to the i-th group,

is the center of group i,

is the nth measurement data. Specifically, the value of i can be set to 4.

Step S3: Using the analysis device 12 to obtain a hard disk performance problem classification model based on the discretized multiple pieces of measurement data and the decision tree algorithm. Among them, the decision tree algorithm is particularly the ID3 algorithm. Further, the decision tree algorithm executed by the analysis device 12 may use one or both of entropy and information gain as an index to measure the importance of features. In addition, the analysis device 12 can receive, for example, the main features most strongly affecting the hard disk given by a professional from the outside through a user interface or a communication interface. In one implementation, the analysis device 12 uses a decision tree algorithm to perform dichotomy on all the discretized measurement data to establish a decision tree, and performs effective verification on the decision tree to ensure that the model has a good learning ability. Proven decision tree as a classification model for hard disk performance problems. In another implementation aspect, obtaining the hard disk performance problem classification model performed by the analysis device 12 includes performing k-fold cross validation. Further, the analysis device 12 divides all the discretized measurement data into k subsets, and takes each of the k subsets as a verification subset and uses the rest as a training subset in turn. where k is in particular 8.

The decision tree algorithm executed in step S3 is exemplarily described below, please refer to FIG. 1 and FIG. 3 , wherein FIG. 3 is a flow chart showing detailed steps of obtaining a hard disk performance problem classification model according to an embodiment of the present invention. . As shown in FIG. 3 , step S3 may include steps S31 to S36 , and steps S31 to S36 are executed by the analysis device 12 .

Step S31: Based on one or both of entropy and information gain, determine a first decision point to divide at least a part of the discretized measurement data into two measurement data groups, wherein the first decision point is associated with in one of the plurality of vibration parameters. Among them, the data matrix divided into two groups by the first decision point can be all discretized measurement data, or (k-1) of the discretized measurement data divided into k subsets Subset. The calculation formulas of entropy and information gain and the judgment threshold can be designed according to actual needs, which are not limited by the present invention.

Step S32: Define i as a positive integer with an initial value of 2.

Step S33: Based on one or both of entropy and information gain, determine the i-th decision point to divide one of the two measurement data groups classified by the (i-1)-th decision point into the other two measurement data sets A dataset wherein the ith decision point is associated with another one of the vibration parameters.

Step S34: Determine whether the value of i is equal to the number of these vibration parameters, wherein the number of these vibration parameters is greater than or equal to two. If it is judged that the value of i is not equal to the quantity of these vibration parameters, execute step S35: add 1 to the value of i, and execute step S33 again; if it is judged that the value of i is equal to the quantity of these vibration parameters, execute step S36: based on the first decision point and execute The results of one or more classification operations form a hard disk performance problem classification model.

Please refer to FIG. 1 and FIG. 4 , wherein FIG. 4 is a binary tree diagram of a decision tree algorithm according to an embodiment of the present invention. The analysis device 12 can establish the decision tree shown in FIG. 4 as a hard disk performance problem classification model by the method described in the above embodiments, wherein the number of vibration parameters is four. The decision tree can include the first to fourth decision points D1~D4 and categories C1~C5, wherein the decision points D1~D4 are respectively associated with different vibration parameters, and the categories C1~C4 are respectively associated with the same vibration as the decision points D1~D4 parameter, while category C5 indicates that it is not related to any of the 4 vibration parameters. Further, each of the decision points D1-D4 classifies the data feed according to a condition associated with a corresponding vibration parameter, wherein the condition indicates a probability range in which the corresponding vibration parameter is the main cause of hard disk performance degradation. The decision tree shown in FIG. 4 is just an example of the distribution of leaf nodes of the decision tree, and theoretically, decision trees with different leaf node distributions may also be generated according to the training data.

For example, decision points D1~D4 are respectively related to resonance frequency, sound pressure, acceleration, and angular acceleration; decision point D1 is based on classification condition c11 "the probability that resonance frequency is the main cause of hard disk performance degradation is less than 0.5" and classification condition c12 "The probability that the resonance frequency is the main cause of hard disk performance degradation is greater than or equal to 0.5" divides the data into decision point D2 or category C1, where category C1 indicates that the main cause of hard disk performance degradation is frequency; decision point D2 is based on classification conditions c21 "The probability that sound pressure is the main cause of hard disk performance degradation is less than 0.5" and the classification condition c22 "The probability of sound pressure is the main cause of hard disk performance degradation is greater than or equal to 0.5" divide the data into decision point D3 or category C2, among them Category C2 indicates that the main cause of hard disk performance degradation is the sound pressure problem; decision point D3 uses the classification condition c31 "the probability that acceleration is the main cause of hard disk performance degradation is less than 0.5" and classification condition c32 "acceleration is the main cause of hard disk performance degradation The probability of the cause is greater than or equal to 0.5" to divide the data into decision point D4 or category C3, wherein category C1 indicates that the main reason for hard disk performance degradation is acceleration; decision point D4 uses the classification condition c41 "angular acceleration is the cause of hard disk performance degradation The probability of the main reason is less than 0.5" and the classification condition c42 "the probability of angular acceleration being the main reason for the performance reduction of the hard disk is greater than or equal to 0.5" to classify the data into category C4 or category C5, where category C4 indicates the main reason for hard disk performance degradation is a spin problem, while category C5 indicates that the main cause of hard drive performance degradation is not related to frequency problems, sound pressure problems, acceleration problems or spin problems.

Please refer to FIG. 5 and FIG. 6 , which are schematic diagrams of the execution environment of a hard disk performance problem analysis method according to an embodiment of the present invention and a flowchart of a hard disk performance problem analysis method according to an embodiment of the present invention. . As shown in Figure 5, the execution environment corresponding to the hard disk performance problem analysis method of the present invention may include a hard disk performance problem classification model establishment system 1 and a computer system 2, wherein the computer system 2 can be connected to the hard disk performance problem classification model by communication Build system 1. The hard disk performance problem classification model establishment system 1 provides a hard disk performance problem classification model as shown in FIG. 1 , and its relevant details have been described in the preceding paragraphs, and will not be repeated here. The computer system 2 includes a processor, the processor can obtain the hard disk performance problem classification model from the hard disk performance problem classification model building system 1, and input the measurement data of the abnormal server hard disk into the hard disk performance problem classification model, so as to Get classification results. The processor is, for example, a microcontroller, a graphics processor, or other electronic devices with data processing functions and storage functions, and is not limited to the scope of the present invention. In this embodiment, the computer system 2 for implementing the hard disk performance problem classification model and the analysis device for establishing the hard disk performance problem classification model are different devices. In another embodiment, the computer system 2 for executing the hard disk performance problem classification model is the same device as the analysis device for establishing the hard disk performance problem classification model.

As shown in Figure 6, the hard disk performance problem analysis method of the present invention may include steps S1~step S4, wherein steps S1~step S3 are the steps for establishing the hard disk performance problem classification model as described in Figure 2, and will not be repeated here Repeat the narrative. Step S4: Using the computer system 2 to input the measurement data of the abnormal server hard disk into the hard disk performance problem classification model, and obtain the classification result, wherein the classification result indicates the ranking among multiple problem factors, and the multiple problem factors are associated with multiple More than one of the vibration parameters. Multiple problem factors may include acceleration problems, rotation problems, sound pressure problems, and resonance frequency problems. Specifically, the computer system 2 can obtain the measurement data of the abnormal server hard disk through the user interface or from the server hard disk measurement device, and use the hard disk performance problem classification model according to the measurement data of the abnormal server hard disk Prioritize the degree of influence of multiple problem factors.

For example, the server has a plurality of hard disk slots, such as 12, and the hard disks connected to 4 of the hard disk slots are abnormal (for example, the performance is not good). After measurement, the first measurement data, The second measurement data, the third measurement data and the fourth measurement data are shown in Table 1. The computer system 2 obtains the first to fourth measurement data through the user interface or from the server hard disk measurement device, and inputs them into the hard disk performance problem classification model to obtain the classification results shown in Table 2 .

Table 1 Acceleration (m/s ² ) Sound pressure (dB) Angular acceleration (rad/s ² ) Frequency (Hz) first measurement data 0.05 93 14.26 300 Second measurement data 0.008 98 9.56 300 The third measurement data 0.14 99 11.22 300 The fourth measurement data 0.24 108 14.6 360

Table 2 Hard Disk Performance Problem Ranking 1 Hard disk performance problem sorting 2 Hard Disk Performance Problem Ranking 3 first measurement data frequency to turn to turn Second measurement data frequency to turn sound pressure The third measurement data frequency to turn to turn The fourth measurement data to turn sound pressure to turn

In summary, the method for establishing a classification model of hard disk performance problems and the system for establishing a classification model of hard disk performance problems of the present invention use the k-average algorithm to discretize the vibration measurement data of multiple single hard disks, and combine Decision tree algorithm can establish a decision tree with high classification accuracy as a hard disk performance problem classification model, and since the vibration measurement data of a single hard disk is used as training data, the established decision tree is pruned The demand is lower. In addition, in the hard disk performance problem analysis method of the present invention, the measurement data of the hard disk with problems inside the servo storage system is input into the above-mentioned hard disk performance problem classification model, and the main reason for the hard disk performance degradation can be predicted well.

In one embodiment of the present invention, the method for establishing a hard disk performance problem classification model, the hard disk performance problem analysis method, and the hard disk performance problem classification model establishment system of the present invention can analyze and test the hard disk loaded on the server, In order to improve the reliability of the server, the server is suitable for artificial intelligence (AI) computing, edge computing (Edge Computing), and can also be used as a 5G server, cloud server or Internet of Vehicles server .

Although the present invention is disclosed by the aforementioned embodiments, they are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, all changes and modifications are within the scope of patent protection of the present invention. For the scope of protection defined by the present invention, please refer to the appended scope of patent application.

1: Hard disk performance problem classification model building system 2: Computer system 11: Vibration parameter sensor 12: Analysis device 111: Three-axis accelerometer 112: Sound pressure meter 113: Resonant frequency analyzer C1~C5: Category c11, c12, c21, c22, c31, c32, c41, c42: classification conditions D1~D4: decision point S1~S4, S31~S36: steps

FIG. 1 is a functional block diagram of a hard disk performance problem classification model building system according to an embodiment of the present invention. FIG. 2 is a flowchart of a method for establishing a hard disk performance problem classification model according to an embodiment of the present invention. FIG. 3 is a flow chart showing detailed steps of obtaining a hard disk performance problem classification model according to an embodiment of the present invention. FIG. 4 is a binary tree diagram of a decision tree algorithm according to an embodiment of the present invention. FIG. 5 is a schematic diagram of an execution environment of a method for analyzing hard disk performance problems according to an embodiment of the present invention. FIG. 6 is a flowchart of a method for analyzing hard disk performance problems according to an embodiment of the present invention.

S1~S3: steps

Claims (8)

A method for establishing a hard disk performance problem classification model, which includes executing with an analysis device: obtaining multiple pieces of measurement data of multiple single hard disks, each of which measurement data includes multiple values of multiple vibration parameters; Discretize the measurement data based on the k-means algorithm; and obtain a hard disk performance problem classification model based on the discretized measurement data and a decision tree algorithm; wherein, based on the discretized measurement data According to the data and the decision tree algorithm, obtaining the classification model of the hard disk performance problem includes: determining a first decision point based on one or both of entropy and information gain, so as to divide the discretized measurement data into At least a part is divided into two measurement data groups, wherein the first decision point is associated with one of the vibration parameters; define i as a positive integer and start as 2, and perform a classification operation, the classification operation includes: Based on one or both of the entropy and the information gain, determining an i-th decision point to divide one of the two measurement data sets classified by the (i-1)-th decision point into the other two measurement data sets a data set, wherein the first decision point is associated with another of the vibration parameters; Judging whether the value of i is equal to the number of these vibration parameters, wherein the number of these vibration parameters is greater than or equal to 2; if the value of i is not equal to the number of these vibration parameters, add 1 to the value of i and perform the classification operation again; And if the value of i is equal to the quantity of the vibration parameters, the hard disk performance problem classification model is formed based on the first decision point and the result of performing one or more classification operations. The method for establishing a hard disk performance problem classification model as described in Claim 1, wherein the decision tree algorithm is ID3 algorithm. The method for establishing a hard disk performance problem classification model according to claim 1, wherein the vibration parameters include multiples of acceleration, angular acceleration, sound pressure, and resonance frequency. The method for establishing a hard disk performance problem classification model as described in Claim 1, wherein based on the discretized measurement data and the decision tree algorithm, obtaining the hard disk performance problem classification model includes performing k-fold cross-validation. A method for analyzing hard disk performance problems, including: using a computer system to input a measurement data of an abnormal server hard disk into the establishment of a hard disk performance problem classification model described in any one of claims 1 to 4 The classification model of the performance problem of the hard disk is established by the method to obtain a classification result; wherein the classification result indicates the ranking among a plurality of problem factors, and the problem factors are related to many of the plurality of vibration parameters. A hard disk performance problem classification model building system, comprising: a plurality of vibration parameter sensors for measuring a plurality of values of a plurality of vibration parameters of each of a plurality of single hard disks; and an analysis device, Connected to the vibration parameter sensors for obtaining multiple measurement data of the single hard disks, discretizing the measurement data based on the k-average algorithm, and discretizing the measurement data based on the discretization and a decision tree algorithm to obtain a hard disk performance problem classification model, wherein each of the measurement data includes the values of the vibration parameters of the corresponding ones of the single hard disks; wherein, based on the discretization Based on the measurement data and the decision tree algorithm, obtaining the hard disk performance problem classification model includes: determining a first decision point based on one or both of entropy and information gain, so that the discretized ones At least a part of the measurement data is divided into two measurement data groups, wherein the first decision point is associated with one of the vibration parameters; define i as a positive integer and start as 2, and perform a classification operation, The classification operation includes: determining an i-th decision point based on one or both of the entropy and the information gain, so as to divide one of the two measurement data groups classified by the (i-1)-th decision point into another two sets of measurement data, wherein the first decision point is associated with another one of the vibration parameters; Judging whether the value of i is equal to the number of these vibration parameters, wherein the number of these vibration parameters is greater than or equal to 2; if the value of i is not equal to the number of these vibration parameters, add 1 to the value of i and perform the classification operation again; And if the value of i is equal to the quantity of the vibration parameters, the hard disk performance problem classification model is formed based on the first decision point and the result of performing one or more classification operations. The system for establishing a classification model for hard disk performance problems as described in Claim 6, wherein the decision tree algorithm is ID3 algorithm. The hard disk performance problem classification model building system as described in claim 6, wherein the vibration parameter sensors include more than one of a three-axis accelerometer, a sound pressure meter and a resonance frequency analyzer, and the three-axis accelerometer is used for One or both of the acceleration and the angular acceleration are measured, the sound pressure meter is used for measuring the sound pressure, and the resonance frequency analyzer is used for measuring the resonance frequency.

Images