CN107832203B - Method for diagnosing rendering performance of mobile terminal - Google Patents

Abstract

The invention discloses a method for diagnosing rendering performance problems of a mobile terminal. And obtaining rendering data of the interface control by modifying the Android platform. And running a large number of application programs on the modified Android platform to obtain a large number of data, and then training a Bayesian network model, wherein the rendering characteristics and the rendering performance indexes are corresponding by the model. And inputting rendering data into the model aiming at a specific interface of a specific application program, calculating a contribution value of the control in the interface to the page rendering performance by using a specific method, and finding out the control which has the largest influence on the interface rendering performance. The method has the advantages that: the rendering performance problem generated by the application program in the mobile terminal can be effectively and accurately diagnosed.

Description

A kind of mobile terminal rendering performance diagnosis method

technical field

The invention relates to a method for diagnosing rendering performance problems of a mobile terminal, in particular to a rendering performance positioning method for terminal applications oriented to application developers.

Background technique

In recent years, as the number of mobile devices and mobile applications has grown rapidly, so has the time users spend on mobile applications. A mobile application produces a response after a user action is called an interaction, and the time required to complete the interaction logic and draw all the interface is called the response time. Typically a user interaction takes a few seconds. Since the interaction response time of mobile applications is directly related to the end-user experience, it has attracted a lot of attention in both academia and industry in recent years. Domestic and foreign researchers have carried out in-depth and extensive research in this area. In recent years, there have been many in-depth studies on interaction response time.

Past research on mobile application response time has mainly focused on heavyweight operations such as network access and IO requests. For example, Applnsight uses binary instrumentation to track the behavior of user actions within the application, and then analyzes the critical execution path to get the root cause of performance problems. Since storage IO accesses cause long delays, a lot of work has focused on measuring and optimizing storage IO access times. For example, H. Kim et al. analyzed the storage system of mobile devices and found that there is a positive correlation between the design of the storage system and the performance of mobile applications, and based on this, they proposed a series of solutions to solve the problems caused by storage IO. performance issues. However, none of these works involve the interface drawing process during the mobile application's responsiveness. In our findings, for a short user interaction process, the time required for interface rendering accounts for a considerable part of the entire interaction response process. Obviously, locating and optimizing performance issues in the rendering process is also one of the keys to improving mobile applications.

SUMMARY OF THE INVENTION

To overcome the above shortcomings of the prior art, the present invention provides a method for locating and diagnosing rendering problems for the rendering process in the user interaction process of the mobile application level.

In order to achieve the above purpose, the technical solution adopted by the present invention is: the mobile terminal performance diagnosis method, comprising the following steps:

1) Modify the Android framework code so that the rendering-related information of each control can be obtained when the application is running.

2) Run a large number of applications on the modified Android platform, collect the rendering information obtained by running these applications, and use these data to construct a Bayesian network.

3) Using the constructed Bayesian network, calculate each control of a specific rendering interface of a specific application, and obtain a rendering performance coefficient (RCS) of the contribution of the control to the rendering time of the entire interface. By sorting all controls by RCS, the controls that have the greatest impact on the rendering performance of a specific interface for a specific application are obtained. Developers can further optimize the application on this basis.

Preferably, the rendering-related data obtained by modifying the Android Framework described in step 1) includes:

1.1 Relevant data of each control. The controls here are the basic elements of interface controls on the Android platform, and all interfaces are composed of multiple controls. This data includes the unique id of the control, the size of the control, and the position of the control.

1.2 The hierarchical relationship of controls. The interface of the Android platform is a tree structure, and there is a hierarchical relationship between the controls. By obtaining the identifier id of the parent control, the hierarchical relationship of all controls in the current interface can be reconstructed.

1.3 Rendering operations and timing. Including the time-consuming of three operations: measure, layout, and draw.

1.4 Data related to user operations. Including the start time of operations such as clicking and sliding, and the FPS frame rate of the system.

Preferably, the construction method of the Bayesian network described in step 2):

2.1 The Bayesian network is a directed acyclic graph, and the nodes of the graph include two types: rendering features rendering features node and performance metric node. The rendering features are represented by f ₁ to f _N , that is, N rendering features; the performance indicators are represented by p ₁ and p ₂ , which are two performance indicators.

2.2 We use corr(fi , f _j ) to represent the correlation coefficient between feature i and feature _j , and use corr(fi , p _j ) to represent the correlation coefficient between feature i and performance index _j , which is expressed in the Bayesian network as Weights of edges between nodes.

2.3 After obtaining the Bayesian network, we further calculate the contribution of the rendering feature i to the performance index j, sig(f _i , p _j ).

Preferably, the calculation method of the rendering performance coefficient RCS described in step 3) is specifically:

3.1 We use con(vi, _fj ) to denote the contribution of a control to feature _j . There are different calculation methods for different features.

3.2 Based on sig(f _i , p _j ) in 2.3 and con(vi , _{f j )} in 3.1, we propose a calculation method to calculate the rendering performance coefficient (RCS) value RCS(vi ₎ for a specific control:

3.3 Using the rendering performance coefficient RCS obtained in step 3.2, we calculate all the control elements of a specific interface and sort them according to the rendering performance coefficient RCS. The higher score has a greater impact on the rendering performance of the interface, and should be the focus of developers. Attention and optimization.

Compared with the prior art, the beneficial effects of the present invention are: the method trains a Bayesian network model through a large amount of data, and corresponds the rendering data of the control with the rendering performance of the final interface; developers can use this method to accurately locate To the UI components that cause the performance of interface rendering to decrease, improve the efficiency of development and problem location.

Description of drawings

Fig. 1 is the method frame diagram of the present invention

Detailed ways

The implementation of a mobile terminal performance diagnosis method of the present invention will be described in detail below with reference to the accompanying drawings, and the steps are as follows:

1. The method of modifying the Android Framework and the rendering-related data obtained are as follows:

1.1. Modifying the Android Framework refers to modifying the source code View.java and Choreographer.java files, and inserting the code output time and related data at the relevant functions. After the modified Framework is compiled, the framework.jar file is generated, uploaded to the Android device through the adb tool, and the original framework.jar is replaced. Restart the device to complete the modification of the Android platform.

1.2. Running on the modified Android platform, the data that can be obtained include the following types:

1.2.1. Data related to each control. The controls here are the basic elements of interface controls on the Android platform, and all interfaces are composed of multiple controls. This data includes the unique id of the control, the size of the control, and the position of the control.

1.2.2. The hierarchical relationship of controls. The interface of the Android platform is a tree structure, and there is a hierarchical relationship between the controls. By obtaining the identifier id of the parent control, the hierarchical relationship of all controls in the current interface can be reconstructed.

1.2.3. Rendering operations and timing. Including the time-consuming of measure, layout, and draw operations.

1.2.4. Data related to user operations. Including the start time of operations such as clicking and sliding, and the FPS frame rate of the system.

2. The construction method of the Bayesian network described in step 2) is as follows:

2.1. Bayesian network is a directed acyclic graph, and the nodes of the graph include two types: rendering features rendering features node and performance metric node. The rendering features are represented by f ₁ to f _N , that is, N rendering features; the performance indicators are represented by p ₁ and p ₂ , which are two performance indicators.

2.1.1. The rendering feature is obtained by processing the interface control rendering data obtained in step 1.2. The specific rendering features are shown in Table 1.

Table 1

2.2. We use corr(fi , f _j ) to represent the correlation coefficient between feature i and feature _j , and use corr(fi , p _j ) to represent the correlation coefficient between feature i and performance index _j , which is expressed in the Bayesian network is the weight of the edge between nodes.

2.3. After obtaining the Bayesian network, we further calculate the contribution of the rendering feature i to the performance index j, sig(f _i , p _j ). The calculation method is to add up the weights of all the edges on the path from the node i to the rendering performance node j, that is, to obtain the contribution value.

3. Calculation method of rendering performance coefficient RCS described in step 3)

3.1. We use con(vi, _fj ) to denote the contribution of a control to feature _j . There are different calculation methods for different features. For example, for the feature "number of empty points", all controls have a value of 1, and for the feature "maximum execution time of drawing operations", the contribution value of the control whose drawing time reaches the maximum drawing time is 1/m, where m is The number of controls whose draw time reaches the maximum draw time; the contribution of controls whose draw time is less than the maximum draw time is 0.

3.2. Based on the rendering feature contribution values sig(f _i , p _j ) in 2.3 and the control feature contribution values con(vi , _{f j )} in 3.1, we propose to calculate the rendering performance coefficient RCS value of a specific control RCS(v _i ) calculation method:

3.3. Using the rendering performance coefficient RCS obtained in 32., we calculate all the control elements of a specific interface and sort them according to the rendering performance coefficient RCS. The higher score has a greater impact on the rendering performance of the interface, and should be used by developers. Focus on attention and optimization.

The content described in the embodiments of the present specification is only an enumeration of the realization forms of the inventive concept, and the protection scope of the present invention should not be regarded as limited to the specific forms stated in the embodiments, and the protection scope of the present invention also extends to the field Equivalent technical means that can be conceived by a skilled person according to the inventive concept.

Claims (1)

1. A method for diagnosing rendering performance of a mobile terminal, including the following steps: 1) Modify the Android framework code so that the rendering-related information of each control can be obtained when the application is running, including: 1.1 The relevant data of each control; the control here is the basic element of the interface control on the Android platform, and all interfaces are composed of multiple controls; these data include the unique id of the control, the size of the control, and the position of the control; 1.2 The hierarchical relationship of controls; the interface of the Android platform is a tree structure, and there is a hierarchical relationship between the controls; by obtaining the identifier (id) of the parent control, the hierarchical relationship of all the controls in the current interface can be reconstructed; 1.3 The operation and timing of rendering; including the time-consuming of three operations: measure, layout, and draw; 1.4 Relevant data of user operations; including the start time of clicking and sliding operations, and the FPS frame rate of the system; 2) Run a large number of applications on the modified Android platform, collect the rendering information obtained by running these applications, and use these data to construct a Bayesian network; the construction method of the Bayesian network is as follows: 2.1 The Bayesian network is a directed acyclic graph. The nodes of the graph include two types: rendering feature renderingfeatures node and performance metric node; rendering features are represented by f ₁ to f _N , that is, N rendering features; performance The indicators are represented by p ₁ and p ₂ , that is, two performance indicators; 2.2 Use corr(fi , f _j ) to represent the correlation coefficient between feature i and feature _j , and use corr(fi , p _j ) to represent the correlation coefficient between feature i and performance index _j , which is expressed as a node in the Bayesian network The weight of the edge between; 2.3 After obtaining the Bayesian network, further calculate the contribution value sig(f _i , p _j ) of the rendering feature i to the performance index j; 3) Using the constructed Bayesian network, calculate each control of a specific rendering interface of a specific application, and obtain a rendering performance coefficient (RCS) of the contribution of the control to the rendering time of the entire interface; Sort to get the controls that have the greatest impact on the rendering performance of a specific interface of a specific application; the calculation method of the rendering performance coefficient (RCS), specifically: 3.1 Use con(v _i , f _j ) to represent the contribution of a control to feature j; there are different calculation methods for different features; 3.2 Based on sig(f _i , p _j ) in 2.3 and con(vi , _{f j )} in 3.1, a calculation method for calculating the rendering performance coefficient (RCS) value RCS(vi ₎ of a specific control is proposed:

3.3 Use the rendering performance coefficient RCS obtained in step 3.2 to calculate all the control elements of a specific interface, and sort them according to the rendering performance coefficient RCS. The higher score has a greater impact on the rendering performance of the interface, and should be paid attention to by developers. and optimization.

Images