TWI781767B - Prediction-based method for analyzing change impact on software components - Google Patents

Abstract

A prediction-based method for analyzing change impact on software components is disclosed. The method comprises the steps of: providing a software system comprising a main software component and at least one auxiliary software component; collecting metrics associated with the workload and each auxiliary software component separately and sequentially before a change of the software system is introduced; calculating correlation coefficients between the collected metrics associated with the workload and each auxiliary software component; if an absolute value of the correlation coefficient is smaller than a threshold value, building a prediction model from the collected metrics associated with the corresponding auxiliary software component; recording metrics associated with the corresponding auxiliary software component sequentially; inputting the collected metrics associated with the corresponding auxiliary software component to the prediction model to obtain predicted metrics of the corresponding auxiliary software component; and calculating a performance difference value by using the recorded metrics.

Description

Prediction-Based Method for Analyzing Performance Impact Due to Content Changes of Software System Components

The present invention relates to a method for analyzing the performance impact due to changes in the content of software system components, and more particularly to a prediction-based method for analyzing the performance impact due to changes in the content of software system components.

When a software system consisting of multiple software components is deployed on a computing device such as a server cluster to meet the requirements of a workload, changes to the software system are often used to improve the performance of the software system. Typical scenarios for changes are software upgrades or adjustments to component configuration parameters. Even when a change is applied to one of the software components, some change effects may inevitably occur and cause performance and/or resource usage ripple effects on other software components in the same application. A key issue for the development operations team of a software system is to understand the impact of that change when the software system introduces that change.

Performance indicators in software application systems, such as memory utilization, CPU utilization, input and output throughput, response time, requests per second, latency, etc., can be monitored, but the real "change impact" is difficult to measure. This is because due to the dynamics and variability of the workload, there is no guarantee that the same workload will be present before and after the change is introduced, so the impact of the change cannot be compared.

A traditional approach for analyzing the performance impact due to changes in the content of software system components is shown in Figure 1. A software application cloud service consisting of four separate software components is deployed to serve a primary workload. Data requests and/or responses exist between software components that are sources of influence for related software components. The cloud service can be an enterprise resource planning (ERP), the main workload comes from an external system, such as a computer mainframe in a factory, and is handled by a software component 1 . Software component 1 handles all operational requests and responses to this external system. Each of the remaining software components supports a specific work function and has internal data requests and responses with other software components. A performance indicator collector installed on the server continuously monitors the performance indicators of all software components. In this case, the performance indicators of the main workload are the performance indicators of software component 1 . If the ERP administrator wants to know how all software components are affected after software component 2 is upgraded, the traditional method can compare the actual operating performance indicators with the current performance indicators, and use the comparison results to check the impact of changes, which can be used to adjust software components configuration parameters or a reference for further upgrades. This is usually accomplished by benchmarking programs. The limitation of this approach is that in order to make meaningful comparisons, it is necessary to find nearly the same workload performance metrics in the before and after changes. Otherwise, the comparison results cannot be trusted because different workload patterns often result in different operational performance metrics for the software components in the system. This is not an easy task in a production environment because the workload changes dynamically. Therefore, this type of benchmark cannot accurately describe the effects of changes introduced into the system.

In order to provide an accurate way to assess the impact of changes to save operating costs, the present invention discloses an innovative method.

This paragraph extracts and compiles certain features of the present invention. Other features will be disclosed in subsequent paragraphs. It is intended to cover various modifications and similar arrangements within the spirit and scope of the appended claims.

According to one aspect of the invention, a prediction-based method for analyzing performance impact due to changes in the content of components of a software system comprises the steps of: a) providing a software system for deployment in a computing hardware environment, the software system comprising satisfying a main software component for requests from a workload, and at least one auxiliary software component for processing specific tasks of the main software component; b) separately and sequentially collecting information about the workload before a change in the software system is introduced with the performance metrics for each ancillary software component; c) calculate a correlation coefficient between the collected performance metrics for the workload and the performance metrics for each ancillary software component; d) if the absolute value of the correlation coefficient is greater than one Threshold, using the collected performance indicators related to the workload and the collected performance indicators related to the corresponding auxiliary software components to establish a prediction model for predicting the corresponding auxiliary software components in a period of time in the future e) have sequentially recorded performance metrics for the corresponding auxiliary software component and the workload during an evaluation period that begins when the change to the software system is introduced; f) input step b) into the predictive model to obtain predicted performance metrics for the corresponding auxiliary software component; and g) use the recorded performance metrics for the corresponding auxiliary software component; The performance indicators of the auxiliary software components and the predicted performance indicators of the corresponding auxiliary software components are used to calculate a performance difference value.

In accordance with another aspect of the invention, a prediction-based method for analyzing performance impact due to changes in the content of components of a software system comprises the steps of: a) providing a software system deployed in a computing hardware environment, the software system comprising a main software component that satisfies a request from a workload, and at least one auxiliary software component that handles a specific job of the main software component; b) separately and sequentially collects information about the job before a change in the software system workload and performance metrics for each auxiliary software component; c) calculate the correlation coefficient between the collected performance metrics for the workload and the performance metrics for each auxiliary software component; d) if the absolute value of the correlation coefficient is less than A threshold value, using the collected performance indicators related to the corresponding auxiliary software component to establish a prediction model for predicting the performance indicator of the corresponding auxiliary software component in a period of time in the future; e) within an evaluation period Sequentially recording the performance indicators about the corresponding auxiliary software component, the evaluation period begins when the change of the software system is introduced; f) inputting the collection of performance indicators about the corresponding auxiliary software component collected in step b) into the predictive model to obtain predicted performance metrics for the corresponding auxiliary software component; and g) using the recorded performance metrics for the corresponding auxiliary software component and the predicted performance metrics for the corresponding auxiliary software component performance index to calculate a performance difference value.

Preferably, the change of the software system can be software system upgrade, software system application configuration parameter adjustment, installation of new auxiliary software components, or deletion of existing auxiliary software components.

Preferably, the computing hardware environment can be a workstation host or server cluster.

Preferably, the performance indicator may be used memory amount, used CPU amount, I/O throughput, response time, requests per second, or latency.

Preferably, the performance difference value may be a mean percentage error (Mean Percentage Error).

Preferably, the collected performance indicators used to build the predictive model fall into two categories.

Preferably, the forecasting model is built by a time series forecasting algorithm.

Preferably, the time series forecasting algorithm is an Auto Regressive Integrated Moving Average (ARIMA) algorithm or a Seasonal Auto Regressive Integrated Moving Average (SARIMA) algorithm.

According to the present invention, the correlation between the performance index of the workload and the performance index of each software component is taken into account, and a prediction model can be established to predict the future development of a certain performance index of a certain software component. Comparing the predicted performance metrics with the actual performance metrics, the impact of changes to said software components can be assessed and the results can be used for further changes and savings in operating costs.

The present invention will be described more specifically by referring to the following embodiments.

Please refer to FIG. 4 first, which illustrates a deployment framework of a software system for analyzing performance impact due to content changes of software system components according to the prediction-based method of the present invention. Figure 4 shows the three operational relationships of software components. The software system includes a main software component A, a first auxiliary software component 1, a second auxiliary software component 2 and a third auxiliary software component 3, and is deployed in a computing hardware environment. The computing hardware environment refers to powerful computing hardware capable of handling complex computing requests from a workload. The computing hardware environment can be, but is not limited to, workstation hosts and server clusters. In this computing hardware environment, there are a large number of central processing units (CPUs), a large number of dynamic random access memory (DRAM) modules (or memory for short), and limited input and output throughput of resources. CPU and DRAM are the resources of the workload, used by the main software component A, which can be subdivided into the actual usage of the first auxiliary software component 1 , the second auxiliary software component 2 , and the third auxiliary software component 3 . The input and output throughput is a comprehensive efficiency value of the computing hardware environment for input and output data. The workload can consume a lot of I/O throughput, and the three auxiliary software components share the same amount of I/O throughput. Similarly, response time, requests per second, and latency are metrics for responsive workloads, with contributions from each auxiliary software component. In the present invention, the performance indicators refer to the amount of memory used, the amount of CPU used, I/O throughput, response time, requests per second, and latency, and are used to analyze the impact of "changes" on all software components. In the embodiment of the present invention, the latency (seconds) of the workload and the number of used CPUs occupied by the auxiliary software components are used for illustration. Changes to software systems may be of different types. For example, it may be software system upgrade, software system application configuration parameter adjustment, installation of new auxiliary software components, deletion of existing auxiliary software components, and so on.

In Figure 4, the main software component A is the element that interacts with the workload in the external system. The performance index of the main software component A is equal to the performance index of the workload. The main software component A receives the request of the workload, executes the corresponding program operation, and returns the response to the specific source of the workload. For example, the workload may be a company's email request, and the main software component A is an email module running on the company's server. According to the present invention, the software system has a technical architecture: in addition to the main software component A that completes the workload request, the software system also has at least one auxiliary software component that handles the specific work of the main software component A. In Fig. 4, the first auxiliary software component 1 "works" directly on the main software component A, and the first auxiliary software component 1 performs data retrieval on all emails. The second auxiliary software component 2 "works" for the first auxiliary software component 1, managing the email content database for all emails. That is, the second auxiliary software component 2 indirectly “works” with the main software component A. The third auxiliary software component 3 "works" for the second auxiliary software component 2 and, under the command of the main software component A, performs data access to an external data center. The request from the main software component A is fulfilled by the first auxiliary software component 1 . There are data between the main software component A and the first auxiliary software component 1, between the first auxiliary software component 1 and the second auxiliary software component 2, and between the second auxiliary software component 2 and the third auxiliary software component 3 (request and responses) are passed.

A performance indicator collector B is also installed in the computing hardware environment, which can be an independent data monitoring software, and is used to collect performance indicators related to software components from each software component. It is emphasized that performance metrics collector B can collect performance metrics about this workload because they are the same as the performance metrics of main software component A.

Please refer to FIG. 2 , which is a flowchart of a prediction-based method for analyzing performance impact due to content changes of software system components according to the present invention. The first step of the prediction-based method is to provide a software system for deployment in a computing hardware environment, the software system comprising a main software component that satisfies a request from a workload, and a software component that handles a specific job of the main software component At least one auxiliary software component (S01). This step is just to define an applicable architecture as described above.

The second step of the prediction-based method is to separately and sequentially collect performance metrics about the workload and each auxiliary software component before a change to the software system is introduced ( S02 ). As mentioned above, the latency associated with the workload and the amount of used CPU occupied by the auxiliary software components are used for illustration. This is using the performance relationship between two different performance metrics to predict the future performance of one of them. In other embodiments, the performance of only one performance indicator is sufficient to predict its future. An example is shown in Fig. 5, which also tabulates the calculated data and results of the correlation coefficients and performance difference values. The performance indicator collector B sequentially collects the performance indicators (delay) of the workload (main software component A) from T1 to T5, and the data are 2, 5, 4, 2, and 3 respectively. The time interval between adjacent time points is the same, for example, 5 seconds, but this is not limited by the present invention, as long as the selected time interval can utilize less hardware resources or have better performance in the analysis of the impact of changes. Can. A change, eg, upgrading the first auxiliary software component 1, occurs at T6. The performance indicator collector B also separately and sequentially collects performance indicators (number of used CPUs) related to the first auxiliary software component 1 , the second auxiliary software component 2 , and the third auxiliary software component 3 from T1 to T5 in sequence. The corresponding data are displayed in the time point column of item description No.2 to No.4.

The third step of the prediction-based method is to calculate the correlation coefficient between the collected performance indicators about the workload and the performance indicators about each auxiliary software component ( S03 ). A correlation coefficient is a numerical measure of some type of correlation between two sets of variables. According to its calculation formula, the correlation coefficient varies between -1 and 1. Taking the data from T1 to T5 of item description No.1 and No.2 to calculate, the correlation coefficient is 0.81. Similarly, taking the data from T1 to T5 of item description No.1 and No.3 to calculate, the correlation coefficient is -0.18. Taking the data from T1 to T5 of item description No.1 and No.4 to calculate, the correlation coefficient is 0.96.

Based on the result of step S03, the prediction-based method has the following different steps. If the absolute value of the correlation coefficient is greater than a threshold, the fourth step is to establish a prediction model based on the collected performance indicators related to the workload and the collected performance indicators related to the corresponding auxiliary software components, for Predict the performance index of the corresponding auxiliary software component in a period of time in the future ( S04 ). Here, thresholds limit the relationship between workloads and each auxiliary software component's hardware resource usage or performance trends. In this example, the threshold is set to 0.7, which indicates that the trend should be very close in the same direction or in the opposite direction, indicating that the performance metrics collected on this workload and the performance metrics collected on the corresponding auxiliary software component have a strong correlation. In other embodiments, the threshold may be any number between 0 and 1, which is not limited in the present invention. As can be seen from FIG. 5, the correlation coefficient between the collected performance index about the workload and the collected performance index about the first auxiliary software component 1, and the collected performance index about the workload and the collected correlation coefficient The correlation coefficient between the performance indicators of the third auxiliary software component 3 meets this requirement. According to the spirit of the present invention, the way of constructing the prediction model is not limited, and any existing data prediction model can be used, even a simple statistical formula. A more accurate predictive model is preferred because it saves resource usage or provides better results. Machine learning predictive models can be used if desired. Preferably, it is established by a time series forecasting algorithm, which may be ARIMA or SARIMA. In this embodiment, the prediction model is established by ARIMA. The prerequisite for establishing the predictive model is that the input must be the performance indicators collected about the workload and the performance indicators collected about the corresponding auxiliary software components before T6. Clearly, the performance metrics collected for building predictive models fall into two categories.

Next, based on the result of step S04, the fifth step of the prediction-based method is to sequentially record performance indicators about the corresponding auxiliary software component and the workload during an evaluation period, the evaluation period starting from the software system when the change is introduced (S05). As mentioned above, the two auxiliary software components, the first auxiliary software component 1 and the third auxiliary software component 3, are the so-called corresponding auxiliary software components in step S05. Therefore, performance indicators related thereto are recorded by performance indicator collector B. The verb "record" used here and the verb "collect" used in step S02 mean the same thing, and they both describe that the performance index collector B acquires data from software components. Different verbs are used in different steps to describe performance metrics. In this embodiment, the evaluation period starts at T6 and ends at T10. The performance indicators related to the workload recorded from T6 to T10 are 1, 3, 7, 2, and 1. There are 5 pieces of performance index data related to the first auxiliary software component 1 or the third auxiliary software component 3 recorded by the performance index collector B. They are 2, 3, 4, 1, and 2 of the first auxiliary software component 1, and 1, 1, 3, 1, and 1 of the third auxiliary software component 3.

The sixth step of the prediction-based method is to input the performance indicators related to the workload and the corresponding auxiliary software components collected in step S02 into the prediction model, so as to obtain the predicted performance indicators of the corresponding auxiliary software components (S06). In FIG. 5 , the input performance indicators related to the workload before the change is applied are 2, 5, 4, 2, and 3, and the input performance indicators related to the first auxiliary software component 1 are 2, 3, 2 , 1, and 2, the input performance indicators related to the third auxiliary software component 3 are 1, 3, 2, 1, and 2. These are collected before the application is applied.

×

其中yi指所有觀察到的數據，x _i是對應到y _i的預測值，而k是估計變量的不同次數。在本實施例中，y _i是在項目說明No.8或No.10，由T6到T10的數字。因此，k為5因為記錄了5組數字。x _i為在項目說明No.11或No.13，由T6到T10的數字。用上面的相關數據計算，記錄的有關於第一輔助軟體組件1的性能指標與預測的第一輔助軟體組件1的性能指標之MPE為50.00%，而記錄的有關於第三輔助軟體組件3的性能指標與預測的第三輔助軟體組件3的性能指標之MPE為-30.00%。 The last step of the prediction-based method is to use the recorded performance indicators for the corresponding auxiliary software component and the predicted performance indicators for the corresponding auxiliary software component to calculate a performance difference ( S07 ). The performance difference value is used to describe the approximate size of the trend and difference between the predicted value and the observed value. There are many methods to generate the performance difference value. In this embodiment, mean percentage error (Mean Percentage Error, MPE) is used. The MPE is the calculated average of the percentage error by which the model's prediction that differs from the actual quantity can be predicted. The formula for MPE is: MPE(x, y)=

x

where yi refers to all observed data, _xi is the predicted value corresponding to _yi , and k is the number of different times to estimate the variable. In this embodiment, y _i is a number from T6 to T10 in item specification No.8 or No.10. Therefore, k is 5 because 5 sets of numbers are recorded. x _i is the number from T6 to T10 in item description No.11 or No.13. Calculated with the relevant data above, the MPE recorded between the performance index of the first auxiliary software component 1 and the predicted performance index of the first auxiliary software component 1 is 50.00%, while the recorded performance index of the third auxiliary software component 3 is 50.00%. The MPE of the performance index and the predicted performance index of the third auxiliary software component 3 is -30.00%.

Please refer to FIG. 6, which is a graph showing the performance metrics related to the workload, the collected/recorded performance metrics related to the first auxiliary software component 1, and the predicted performance metrics of the first auxiliary software component 1 over time. chart. Before T6, the trend of the workload is similar to the trend of the performance index collected about the first auxiliary software component 1 , with peaks and troughs appearing at the same point in time. A prediction (shown as the dotted line) is obtained following the steps above. The recorded performance index with respect to the first auxiliary software component 1 is different from the predicted performance index of the first auxiliary software component 1 and has a different trend. On average, the changes caused the predicted performance indicators of the first auxiliary software component 1 to be 50.00% higher than they should be. Similarly, see FIG. 8 , which shows performance metrics related to the workload, collected/recorded performance metrics related to the third auxiliary software component 3, and predicted performance metrics of the third auxiliary software component 3 over time. Time-varying graph. Prior to T6, the trend of the workload is similar to the trend of the performance metrics collected for the third auxiliary software component 3 . A prediction is also obtained according to the above steps (as shown by the dotted line). The recorded performance indicators related to the third auxiliary software component 3 and the predicted performance indicators of the third auxiliary software component 3 are different and have different trends. On average, the changes caused the predicted performance indicators of the third auxiliary software component 3 to be 30.00% lower than they should be. Once the performance difference value is obtained, the impact amount of the performance index caused by the change can be predicted, and necessary adjustments can be made to the computing hardware environment.

Under the condition that the absolute value of the correlation coefficient is less than the threshold, the present invention has another way to analyze the performance impact caused by the content change of the software system components. Please refer to FIG. 3 , which is another flowchart of a prediction-based method for analyzing performance impact due to content changes of software system components according to the present invention.

If the absolute value of the correlation coefficient is less than a threshold, the fourth alternative step is to establish a prediction model based on the collected performance indicators related to the corresponding auxiliary software component to predict the corresponding auxiliary software component in a future period of time. Performance indicators of auxiliary software components (S04'). Here, the threshold remains unchanged at 0.7. An absolute value of the correlation coefficient less than 0.7 also indicates a weak or no correlation between the performance metrics collected about the workload and the performance metrics collected about the corresponding auxiliary software component. It can be seen from FIG. 5 that the correlation coefficient between the collected performance index related to the workload and the collected performance index related to the second auxiliary software component 2 meets this requirement. The forecasting model was built by ARIMA. The prerequisite for establishing the predictive model is that the input must be the performance index related to the second auxiliary software component 2 collected before T6.

Then, based on the result of step S04', an alternative fifth step of the prediction-based method is to sequentially record performance indicators for the corresponding auxiliary software component during an evaluation period that begins with the software system's When the change is introduced (S05'). Here, the second auxiliary software component 2 is the so-called corresponding software component in step S05'. Therefore, performance indicators related to the second auxiliary software component 2 are recorded by the performance indicator collector B. The performance indicators related to the first auxiliary software component 2 recorded from T6 to T10 are 3, 2, 3, 2, and 3.

The alternative sixth step of the prediction-based method is to input the performance index about the corresponding auxiliary software component collected in step S02 into the prediction model, so as to obtain the predicted performance index of the corresponding auxiliary software component ( S06'). In Figure 5, the input performance indicators are 2, 2, 3, 3, and 4.

The alternative final step of the prediction-based method is to use the recorded performance indicators for the corresponding auxiliary software component and the predicted performance indicators for the corresponding auxiliary software component to calculate a performance difference value (S07 '). Step S07 is exactly the same as step S07', but the calculation data is generated in a different way. MPE is still used as a performance differential value. According to this formula, y _i is the number from T6 to T10 in Item Description No.9. k is 5. x _i is the number from T6 to T10 in item description No.11 or No.13. Calculated with the above relevant data, the MPE between the recorded performance index of the second auxiliary software component 2 and the predicted performance index of the second auxiliary software component 2 is 90.00%.

Please refer to FIG. 7, which is a graph showing the performance metrics related to the workload, the collected/recorded performance metrics related to the second auxiliary software component 2, and the predicted performance metrics of the second auxiliary software component 2 over time. chart. Before T6, the trend of the workload is not similar to the trend of the collected performance indicators for the second auxiliary software component 2 . A prediction (shown as the dotted line) is obtained according to the above alternate steps. The recorded performance index with respect to the second auxiliary software component 2 is different from the predicted performance index of the second auxiliary software component 2 and has a different trend. On average, the changes resulted in 90.00% of the predicted performance indicators of the second auxiliary software component 2 being higher than they should be.

In this embodiment, the time points successively come one after another. In practice, there can be an interruption between T5 and T6. That is, data for building predictive models can be collected earlier in the introduction of changes. Also, because workload patterns are likely to be based on time of day or day of week, it is beneficial for software systems to build predictive models at similar times of day (or day of week) for each analysis. Collected/recorded performance metrics may be obtained at other times.

Analysis of the impact of changes has the following advantages. First, the software component affected by the change and its affected value can be identified. For the DevOps team, they want to know if one or more software components change after a new software release, the performance of the software system is gained or lost, and the engineering team can confirm whether the results are as expected or if there are any abnormalities. It's feedback to the engineering team. Second, it is easy to assess whether certain system parameters should be tuned for this change. For example, configuration settings for database/backend services may add a new cluster node, several CPUs or memory modules to the computing hardware environment. Operations teams can also analyze quantifiable results to help them assess whether the changes they made met their expectations. If the performance impact is too great, one possible course of action might be to roll back the changes made.

Although the present invention has been disclosed above in terms of implementation, it is not intended to limit the present invention. Anyone with ordinary knowledge in the technical field may make some changes and modifications without departing from the spirit and scope of the present invention. Therefore, The scope of protection of the present invention should be defined by the scope of the appended patent application.

1: The first auxiliary software component 2: The second auxiliary software component 3: The third auxiliary software component A: Main software component B: Performance indicator collector S01~S07: Steps S04'~S07': steps

FIG. 1 depicts a software system deployment framework for a traditional method of analyzing performance impact due to content changes of software system components.

2 is a flowchart of a prediction-based method for analyzing performance impact due to changes in the content of software system components in accordance with the present invention.

3 is another flowchart of a prediction-based method for analyzing performance impact due to changes in software system component content in accordance with the present invention.

FIG. 4 illustrates a software system deployment framework for analyzing performance impact due to content changes of software system components according to the prediction-based method of the present invention.

Figure 5 tabulates the calculated data and results of the correlation coefficients and performance difference values.

6 is a graph showing performance metrics related to a workload, collected/recorded performance metrics related to a first auxiliary software component, and predicted performance metrics of the first auxiliary software component over time.

7 is a graph showing performance metrics related to a workload, collected/recorded performance metrics related to a second auxiliary software component, and predicted performance metrics of the second auxiliary software component over time.

8 is a graph showing performance metrics related to a workload, collected/recorded performance metrics related to a third auxiliary software component, and predicted performance metrics of the third auxiliary software component over time.

S01~S07: Steps

Claims (9)

A prediction-based method for analyzing performance impact due to changes in the content of components of a software system comprising the steps of: a) providing a software system for deployment in a computing hardware environment, the software system including components that satisfy requests from a workload a main software component, and at least one auxiliary software component that handles a specific job of the main software component; b) collecting information about the workload and each auxiliary software component separately and sequentially prior to the introduction of a change to the software system performance index; c) calculating the correlation coefficient between the performance index collected about the workload and the performance index about each auxiliary software component, wherein the correlation coefficient is a numerical measure of the correlation between two sets of variables, the correlation coefficient between -1 and 1; d) if the absolute value of the correlation coefficient is greater than a threshold, the collected performance indicators on the workload and the collected performance indicators on the corresponding auxiliary software components establishing a prediction model for predicting the performance index of the corresponding auxiliary software component in a period of time in the future; e) sequentially recording the performance index of the corresponding auxiliary software component and the workload during an evaluation period, The evaluation period begins when the change to the software system is introduced; f) inputting the performance metrics collected in step b) about the workload and the corresponding ancillary software components into the predictive model to obtain the predicted the performance index of the corresponding auxiliary software component; and g) using the recorded performance index for the corresponding auxiliary software component and the predicted performance index for the corresponding auxiliary software component to calculate a performance difference value. A prediction-based method for analyzing performance impact due to changes in the content of software system components comprising the steps of: a) providing a software system deployed in a computing hardware environment, the software system comprising a main software component that satisfies a request from a workload, and at least one auxiliary software component that handles specific tasks of the main software component; b) Before a change in the software system is introduced, collect performance metrics about the workload and each ancillary software component separately and sequentially; c) calculate the collected performance metrics about the workload and about each ancillary software component The correlation coefficient between the performance indicators of the components, wherein the correlation coefficient is a numerical measure of the correlation between two groups of variables, and the correlation coefficient is between -1 and 1; d) if the absolute value of the correlation coefficient is less than a threshold value, take These collected performance indicators related to the corresponding auxiliary software component establish a prediction model for predicting the performance indicator of the corresponding auxiliary software component in a period of time in the future; e) sequentially record during an evaluation period With respect to the performance index of the corresponding auxiliary software component, the evaluation period begins when the change in the software system is introduced; f) inputting the performance index collected in step b) with respect to the corresponding auxiliary software component into the predictive model to obtain predicted performance metrics for the corresponding auxiliary software component; and g) using the recorded performance metrics for the corresponding auxiliary software component and the predicted performance metrics for the corresponding auxiliary software component to Compute a performance difference value. The prediction-based method according to claim 1 or 2, wherein the change of the software system is software system upgrade, software system application configuration parameter adjustment, installation of new auxiliary software components, or deletion of existing auxiliary software components. The prediction-based method according to claim 1 or 2, wherein the computing hardware environment is a workstation host or a server cluster. The prediction-based method as described in claim 1 or 2, wherein the performance indicator is the amount of used memory, the amount of used CPU, input and output throughput, response time, number of requests per second, or delay. The prediction-based method according to claim 1 or 2, wherein the performance difference is a mean percentage error (Mean Percentage Error). The prediction-based method according to claim 1 or 2, wherein the collected performance indicators used to build the prediction model are divided into two categories. The forecast-based method of claim 1 or 2, wherein the forecast model is established by a time series forecast algorithm. The method based on forecasting as described in claim item 8, wherein the time series forecasting algorithm is a difference integrated moving average autoregressive (Auto Regressive Integrated Moving Average, ARIMA) algorithm or a seasonal difference integrated moving average autoregressive (Seasonal Auto Regressive Integrated Moving Average, SARIMA) algorithm.

Images