KR20060063236A - Analysis method of distributed application transaction performance based on analysis of traffic data file - Google Patents

Abstract

The present invention relates to a method and system for statistically analyzing the performance of an application transaction operating in a distributed environment, the method comprising: generating basic data for analyzing an application transaction performance by analyzing a traffic capture file;

Generating application time relational transaction performance report by relational statistics.

According to the present invention, it is possible to obtain statistical information on transactions and related matters that can quantitatively check application performance in a distributed application operating environment, thereby enabling a quantitative approach in application system optimization, capacity planning, and performance problem analysis. .

Packet Analysis, Performance Statistics, Transactions, Capacity Planning, Problem Analysis, Optimization, Quantitative, Quantitative

Description

ANALYSIS METHOD OF DISTRIBUTED APPLICATION TRANSACTION PERFORMANCE BASED ON ANALYSIS OF TRAFFIC DATA FILE}

1 is a block diagram showing a transaction response time analysis list according to the prior art.

Figure 2 is a conceptual diagram showing a connection state of the analysis system according to an embodiment of the present invention.

3 is a structural diagram showing components of an analysis system;

4 is a conceptual diagram illustrating a state in which a transaction response time analysis list is divided into fixed parts and variable parts.

5 is a conceptual diagram illustrating a state in which a kind identifier is assigned to each transaction.

6 is a performance statistics table based on the number of performance analysis by analyzing the transaction response time in the analysis method according to an embodiment of the present invention.

7 is a performance statistics table as a response time histogram by analyzing the transaction response time in the analysis method according to an embodiment of the present invention.

The present invention relates to a method for analyzing distributed application transaction performance based on traffic data file analysis. More specifically, the application performance is statistically analyzed by analyzing a file (hereinafter, a 'capture file') that collects traffic flowing through a network in a distributed application environment. The present invention relates to a method for analyzing distributed application transaction performance based on traffic data file analysis.

1 is a block diagram showing a transaction response time analysis list according to the prior art. As shown in FIG. 1, the method of preparing application performance analysis data through distributed application traffic analysis according to the related art includes names of all transactions or threads (hereinafter, referred to as “transactions”) performed during a unit time. The output was generated by simply listing each execution time (response time, 1020), traffic bytes, and the like. This method required a lot of system resources of the analysis computer because it had to output on the screen all transactions performed regardless of the type and amount of transactions performed during the unit time.

In addition, because too many transactions are listed on the screen, it was difficult for the user to identify a transaction that had a problem with its performance, as well as the overall performance of the homogeneous transaction, the impact of the performance aspect of the transaction on the system, and the overall system performance. Quantitative figures based on statistical processing, which is the goal of application performance analysis, such as performance, were not available.

That is, the transaction of "SELECT COL_A FROM TAB_A WHERE COL_A = 123" indicated by identifier 1 and "SELECT COL_A FROM TAB_A WHERE COL_A = 10" indicated by identifier 4 are homogeneous transactions, but are not distinguished by homogeneous transactions, but are listed as different transactions. Done.

Therefore, in order to analyze distributed application transactions using traffic analysis method, and to understand the performance more quickly and accurately, a method and system for preparing a product by a method of statistically calculating response time by classifying individual transactions by type are required.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and analyzes the distributed application transactions performed for each unit time by type and analyzes the processing time of the transaction statistically in a time-relational manner, and the overall application performance and system usage situation. The purpose of this paper is to provide a distributed application transaction performance analysis method based on traffic data file analysis that can provide a quantitative understanding.

In addition, the present invention is based on the analysis of traffic data files, which allows access to a number of tasks that previously had to rely on experience and estimation in a quantitative and arithmetic way, such as prioritizing the optimized transactions, predicting the effects of optimization, and capacity planning Its purpose is to provide a distributed application transaction performance analysis method.

Disclosed is a distributed application transaction performance analysis method based on traffic data file analysis, which captures and stores traffic data flowing before or after an application server through a network, and extracts a transaction and response time. Making a step; Dividing the transaction into a fixed part and a variable part, and storing a transaction having the same fixed part as being the same transaction; Storing information for each transaction, the number of executions that are the sum of the number of responses to the same transaction and the response time that is the sum of the time taken to respond to the same transaction; Calculating a total number of executions, which is the sum of the number of executions for all transactions, and calculating a ratio of the number of executions for each transaction to the total number of executions; Calculating a total response time that is a sum of response times for all transactions, and calculating a ratio of response time for each transaction to the total response time; Calculating an average response time which is a value obtained by dividing the response time of each transaction by the number of executions of each transaction.

The fixed part of the transaction is a common execution part that is commonly performed regardless of the execution environment of the unit module, and means a part for determining the purpose of the execution content, and the variable part of the transaction is an input argument or This is the part corresponding to the parameter, and it means the part that determines the result of the execution contents.

Among the information on the plurality of response times stored for each transaction, storing information on at least one of the minimum response time, the maximum response time, the intermediate response time, and the least response time for each transaction; do.

The minimum response time is the shortest response time among the plurality of response times for each transaction, and the maximum response time is the longest response time among the plurality of response times for each transaction.

The intermediate response time is an arithmetic mean value of the minimum response time and the maximum response time for each transaction, and the least response time is a response time that appears most frequently among a plurality of response times for each transaction.

The performance statistics table is prepared by arranging the number of times of execution, response time, average response time, minimum response time, maximum response time, intermediate response time, least frequent response time, and transaction name for each transaction in the order of execution frequency or response time. Storing; further comprises.

Calculate the average response time, minimum response time, maximum response time, intermediate response time and mode response time of each transaction from the average response time, minimum response time, maximum response time, intermediate response time, and mode response time for each transaction. Storing in the performance statistics table, wherein the average response time, the minimum response time, the maximum response time, the intermediate response time, and the least frequent response time of each transaction are the average response time, the minimum response time, The maximum response time, the intermediate response time, and the most frequent response time are multiplied by the number of executions for each transaction, and then summed, and the sum is divided by the number of executions of the entire transaction.

Hereinafter, the present invention will be described with reference to the drawings.

FIG. 2 is a conceptual diagram showing a connection state of a distributed application transaction performance analysis system (hereinafter referred to as an 'analysis system') based on traffic data file analysis according to an embodiment of the present invention, and FIG. 3 is a component of an analysis system. It is a structural diagram showing.

Referring to Figures 2 and 3, the analysis system 100 is conceptually installed outside the application server 200.

Before or after the application server 200, a plurality of user terminals 400 and the database system 300 are each connected to the network.

The application server 200 is a server of a system that performs various commands (transactions) input from the user terminal 400 and provides a result value to the user.

The database system 300 is a system for storing various data required for the application server 200 to provide search data.

The analysis system 100 includes a central processing unit 110, a capture file DB 120, a response time information DB 130, and performance statistics information DB 140.

The central processing unit 110 captures and stores a transaction packet as a file, detects and stores a transaction and each response time from the stored traffic capture file, and generates performance statistics by analyzing the response time for each transaction.

The capture file DB 120 is a database for capturing and storing traffic flowing in a network section before or after the application server 200.

The response time information DB 130 stores the application server 200 or the database system 300 together with all the individual transactions that have performed the elapsed time until the data processing request is received and the result data is returned. do. That is, the identifier 1010, the response time 1020, the transaction name 1030, and the like of each transaction are stored for each field.

The performance statistics information DB 140 stores performance statistics information created by analyzing the information on the response time stored in the response time information DB 130.

4 is a conceptual diagram illustrating a state in which a transaction response time analysis list is divided into fixed parts and variable parts, and FIG. 5 is a conceptual diagram illustrating a state in which a type identifier is assigned to each transaction.

First, the central processing unit 110 analyzes the traffic capture file collected before or after the application server 200 in a manner similar to the conventional art, and creates a transaction response time list as shown in FIG. 1. As described above, the transaction response time list includes a transaction identifier 1010, a response time 1020, which is a time taken to perform a transaction, and a transaction 1030 actually performed for each transaction.

The central processing unit 110 classifies transactions. In order to classify the transactions generated by the client program by type, it is necessary to recognize only a portion capable of identifying the type of transaction in the transaction data generated during execution. To this end, the central processing unit 110 divides each transaction into a fixed portion 1032 and a variable portion 1034 as shown in FIG. 4, and assigns a separate identifier for identifying the type of transaction to the fixed portion 1032. . Through this process, each transaction can be classified by type.

In the present invention, the fixed part 1032 and the variable part 1034 are used as a criterion for classifying transactions. The fixed part 1032 means a common execution part that is commonly performed regardless of the execution environment when performing a unit module operation. It determines the purpose of the contents. In other words, it is a part of a program that is coded in a program and remains the same even if the user or execution environment is changed.

On the contrary, the variable part 1034 is a part corresponding to an input argument or parameter, and is a part which determines the result of the execution content. For example, when a transaction is input to search for the name 'Kim Chul Soo', the 'search command' part may be regarded as a fixed part 1032 and the name 'Kim Cheol Soo' may be regarded as a variable part 1034.

The type identifier 1040 shown in FIG. 5 is a mark for identifying the fixed portion 1032 of each transaction. The reason for providing a separate type identifier 1040 is not to directly use the fixed portion 1032 as an identifier. This is because the processing speed is advantageous.

Figure 5 illustrates the results of this process and shows that the same type of transaction has the same transaction type identifier 1040. Once the type identifier 1040 is assigned as shown in FIG. 5, performance statistics of each transaction can be obtained simply.

6 is a performance statistics table based on the number of performance analysis by analyzing the transaction response time in the analysis method according to an embodiment of the present invention. The central processing unit 110 creates a performance statistics table from the identifier 1010, the response time 1020, and the transaction name 1030 of each transaction stored in the response time information DB 130, and stores the performance statistics table in the performance statistics DB 140.

As shown in FIG. 6, the transaction performance statistics table is divided into a name unit 2300 indicating a transaction type, a performance count statistics unit 2100, and a response time statistics unit 2200, and the execution times and response time statistics units 2100 and 2200. Are again classified into their respective statistics sections. Here, the type of transaction is that the type identifier 1040 given in FIG. 5 views and classifies the same transaction as the same transaction.

The performance recovery statistics unit 2100 is divided into a performance recovery rank 2102, a performance recovery total 2104, a performance recovery ratio 2106, and a cumulative performance recovery ratio 2108.

In addition, the response time statistics unit 2200 may include a response time rank 2202, a response time total 2204, a response time ratio 2206, a cumulative response time ratio 2208, a minimum response time 2210, and an average response time 2212. ), An intermediate response time 2214, a mode response time 2216, and a maximum response time 2218.

The number of times of execution 2102 in the number of execution statistics 2100 is to arrange each transaction in the order of the highest number of times of execution of transactions, and the total number of times of execution 2104 is the number of times of execution of each transaction. In addition, the execution frequency ratio 2106 represents a ratio occupied by a specific transaction among the total transactions executed by the application server 200, and the cumulative execution ratio 2108 is a cumulative number of executions accumulated from a transaction of the maximum execution frequency. Indicates a ratio.

As shown in FIG. 6, since the total number of executions of 1522 transactions is 28,524 and the total number of executions 2104 of the second-ranked transaction (GET / buy / top_detail_view.asp) is 3,515, 3,515 / 28,524 = 12.323 ( %) Of the performance recovery rate (2106).

On the other hand, in the response time statistics unit 2200, the response time rank 2202 is a rank according to the total sum of the response time, the response time sum 2204 is the total response time required to perform the transaction, and the response time ratio 2206. ) Represents the ratio of the response time of a specific transaction to the total response time of the application server 200 as a whole. Also, the minimum response time 2210 and the maximum response time 2218 represent the minimum and maximum response time in a specific transaction, and the average response time 2212 is an arithmetic mean of the total response time for a specific transaction, and an intermediate response time ( 2214 denotes a median value, and the mode response time 2216 indicates a response time represented by the maximum number of times.

The minimum response time 2210 and the maximum response time 2218 of the entire application server 200 are inconvenient to briefly grasp the situation of the system in the case of simply indicating the shortest or longest time. It is not suitable to do this. Therefore, the minimum response time 2210 and the maximum response time 2218 of the entire system need to be calculated in consideration of weights according to the response times of each transaction. In the present invention, the minimum response time 2210 and the maximum response time 2218 of the entire system are calculated by the following method.

or

For example, the total minimum response time (0.076) shown in FIG. 6 is multiplied by the total number of times of execution (3,542) by the minimum response time (0.024) of the first priority transaction, and then the minimum response time (0.000; below the decimal point) of the second priority transaction. There is a non-zero number, so multiplying is not 0) and multiplying the total recovery times (3,515) to get the values of all transactions, then sum the total divided by the total number of times (28,524). The same may be applied to obtaining the total maximum response time (3.986), as well as the average response time (2212), the intermediate response time (2214), the least response time (2216) of the overall analysis system 100.

As shown in FIG. 6, the response time total 2204 of “GET / buy / top_detail_view.asp”, which is the seventh response time rank 2202, is 394.424 seconds, and 2.391 for the total response time of the application server 200 16,496.948 seconds. It can be seen that it has a response time ratio 2206 of (%). In addition, since 3,515 transactions were performed in 394.424 seconds, the average response time was 394.424 / 3,515 = 0.112 seconds.

In addition, the 'GET /buy/detail_view.asp' transaction, which is the No. 1 based on the number of executions of FIG. 6, accounts for 12.418 percent of the total executions, 47.439 percent in terms of response time, and the arithmetic average response time takes 2.209 seconds. Can be understood immediately. Furthermore, the improvement effect of optimizing the average response time of 2.209 seconds to average one second can be calculated simply based on the response time ratio of 47.439 percent of the total response time. In addition, the ratio of the response time to the total can be regarded as the effect on the performance of the entire system, thereby quantitatively understanding the importance evaluation and the priority of optimization.

Meanwhile, in FIG. 6, the ranking of each transaction is arranged based on the number of executions, but the ranking may be displayed by changing the criteria according to the response time or the average time according to the needs of the user.

7 is a performance statistics table represented by the response time histogram by analyzing the transaction response time in the analysis method according to an embodiment of the present invention. Using these histograms, it is possible to express not only numerically but also visually the types of transactions that can be optimized and the expected effects of improvement. For example, the first item of FIG. 7 shows that the number of transaction execution times (3014) having a response time of 3 seconds or more is 377, and the total of these 377 response times is 7.922 percent of the total execution time.

Although the embodiments of the present invention have been described above, the scope of the present invention is not limited to these embodiments, and the right also extends to those skilled in the art.

According to the present invention, it is possible to identify priorities of optimization targets, to predict the optimization effect, and to quantitatively access capacity planning for transactions used in a distributed application environment. The problem analysis work in the abnormal state can be performed quickly.

In addition, the rank and cumulative number of execution times and response time shown in the transaction performance statistics table indicate the actual user usage behavior, and when applied to the load test, more accurate load test is possible.

Claims (7)

In the distributed application transaction performance analysis method, Capturing and storing traffic data flowing before or after the application server through the network, and extracting a transaction and response time; Dividing the transaction into a fixed part and a variable part, and storing a transaction having the same fixed part as being the same transaction; Storing information for each transaction, the number of executions that are the sum of the number of responses to the same transaction and the response time that is the sum of the time taken to respond to the same transaction; Calculating a total number of executions, which is the sum of the number of executions for all transactions, and calculating a ratio of the number of executions for each transaction to the total number of executions; Obtaining a total response time that is a sum of response times for all transactions, and calculating a ratio of response time for each transaction to the total response time; And calculating an average response time obtained by dividing the response time of each transaction by the number of executions of each transaction. The method of claim 1, The fixed part of the transaction It is a common execution unit that is performed in common regardless of the execution environment when performing unit module work, and means a part that determines the purpose of the execution contents. The variable part of the transaction A method for analyzing distributed application transaction performance based on traffic data file analysis, which is a part corresponding to an input argument or parameter, and means a part for determining a result of an execution content. The method of claim 1, Among the information on the plurality of response times stored for each transaction, storing information on at least one of the minimum response time, the maximum response time, the intermediate response time, and the least response time for each transaction; Distributed application transaction performance analysis method based on traffic data file analysis. The method of claim 3, The minimum response time The shortest response time among multiple response times for each transaction. The maximum response time is Distributed data transaction performance analysis method based on traffic data file analysis, characterized in that the longest response time of the plurality of response time for each transaction. The method of claim 3, The intermediate response time Is the arithmetic mean of the minimum and maximum response times for each transaction. The least response time is Distributed data transaction performance analysis method based on traffic data file analysis, characterized in that the response time that appears most frequently among a plurality of response times for each transaction. The method of claim 3, The performance statistics table is prepared by arranging the number of times of execution, response time, average response time, minimum response time, maximum response time, intermediate response time, least frequent response time, and transaction name for each transaction in the order of execution frequency or response time. Distributed data transaction performance analysis method based on the traffic data file analysis further comprising the step of storing. The method of claim 6, Calculate the average response time, minimum response time, maximum response time, intermediate response time and mode response time of each transaction from the average response time, minimum response time, maximum response time, intermediate response time, and mode response time for each transaction. Storing in the performance statistics table; The average response time, the minimum response time, the maximum response time, the intermediate response time, and the most frequent response time of the entire transaction are the average response time, the minimum response time, the maximum response time, the intermediate response time, and the least frequent response time for each transaction. Distributed data transaction performance analysis method based on traffic data file analysis, characterized in that by multiplying the number of executions per transaction of the sum and then dividing the sum by the number of executions of the entire transaction.

Images