WO2004075061A1

WO2004075061A1 - System performance measurement/analysis device

Info

Publication number: WO2004075061A1
Application number: PCT/JP2004/002011
Authority: WO
Inventors: Myongho Ahn; Masami Takai; Atsuko Yamada
Original assignee: Intellasset, Inc.
Priority date: 2003-02-24
Filing date: 2004-02-20
Publication date: 2004-09-02
Also published as: JP2004264914A

Abstract

Applications of the system whose performance is to be measured are analyzed and the processing unit constituting the applications is extracted. Correlation between the processing units is clarified and the hardware resource constituting the system on which each processing unit is operating is clarified. Moreover, the operation performance of each processing unit (time required for execution, communication time between each processing unit, and the like) is measured without modifying the application source code. In addition, at processing execution stage, a load state of each hardware resource is measured and displayed together with the operation performance of each processing unit, hardware resource operating, and the hardware resource load state.

Description

Description System performance measurement analyzer Technical field

The present invention optimizes the cooperative operation of a system by measuring and analyzing the operating performance of the entire system in a system environment configured by a network, a plurality of servers and software connected by a network. On how to do. Background art

Nowadays, the use of computers has become very popular, and computers are used everywhere. In addition, it has become common to connect multiple computers (servers) over a network and distribute and use various applications on those servers. In the past, these applications were often used individually, but with the spread of the Internet and the Web, the use of multiple applications in cooperation to perform desired processing has increased today. Web services is a typical example. In such a cooperative operation environment of applications, since a plurality of applications are involved in a desired process, in order to improve the operation performance of the desired process, all applications involved in the desired process ( It is necessary to improve the operation performance of these applications as a whole. Conversely, if the performance of even one of the applications in the system deteriorates, it becomes a bottleneck in throughput, and the operating performance of the system deteriorates. In such an environment, all applications and IT resources in the system are monitored, and factor analysis and solution presentation to eliminate bottlenecks in system performance (this is used to optimize the system) ) Is required.

Conventionally, when performance degradation of application execution has occurred in a computer system, the following method has been used to identify the cause.

1. Measure the load status of hardware resources (CPU, hard disk, memory, network card, etc.) and identify hardware resources with high load.

2. Measure the individual processing performance of the application group and identify the application where performance degradation is observed.

Patent Literature 1 and Patent Literature 2 discuss conventional approaches of a system for monitoring the load status of IT resources. Patent Document 3 discloses a system for monitoring the performance of software without depending on the language. Patent Document 4 discloses a method of visualizing and displaying a software configuration, and Patent Document 5 discloses a method of analyzing a source code.

[Patent Document 1]

US Patent 5,572,672

〖Patent Document 2]

U.S. Patent 5,506,955

[Patent Document 3]

U.S. Application Publication US 2002/0095660

[Patent Document 4]

US Patent No. 6,226,787

[Patent Document 5]

US Patent 5,500,881

In a traditional system environment where applications have been deployed and run on a single piece of hardware, a combination of the above methods can result in the degradation of which application is caused by which hardware source. It is relatively easy to determine if you have been.

In recent years, the number of applications that are distributed and deployed on multiple pieces of hardware connected by a network has been increasing. In such an environment, since one application can be executed on multiple hardware resources, it is difficult to make one-to-one correspondence between the application and the hardware resource executing the application.

To solve this problem, the application is divided into smaller processing units, the performance of each processing unit is evaluated, and performance degradation occurs in both the processing unit and the hardware resources on which it runs. Factors need to be identified. Disclosure of the invention

It is an object of the present invention to provide a system performance measuring and analyzing apparatus capable of measuring the operation performance of an application in more detail and providing appropriate information for optimizing the cooperative operation.

In addition, in the conventional method for measuring the operation performance of an application, it is assumed that the source code of the application is obtained or modified, but the present invention also has a problem that it is unnecessary.

The system performance measurement / analysis apparatus of the present invention is installed on a plurality of servers connected via a network, and a part or all of the plurality of applications cooperate in a virtual machine environment. Is a system performance measurement and analysis device that measures and analyzes the results and optimizes the cooperative operation based on the results.

Application analysis means for analyzing and extracting processing units constituting the application, and analyzing and acquiring a calling relationship between the processing units;

Operating performance measuring means for measuring the operating performance of the processing unit; Hardware resource specifying means for analyzing and acquiring a correspondence relationship between the processing unit and hardware resources used by the processing unit;

Display means for displaying the calling relationship between the processing units, the operation performance of the processing unit, and the hardware resources on which the processing unit operates together (this is referred to as an object / destruction diagram). And BRIEF DESCRIPTION OF THE FIGURES

FIG. 1 is a diagram showing a configuration of .NET which is one example of a virtual machine environment.

Figure 2 shows the concept of generating a binary file from source code and restoring source code from the binary file.

Figure 3 shows the flow of the source code restoration process.

Figure 4 shows the flow of the process of the preparation process of the lazy worker method.

Figure 5 is a flow chart of the lazy generation process. FIG. 6 is a diagram showing a system configuration of the device according to the embodiment of the present invention. FIG. 7 is a conceptual diagram of a functional configuration of software according to the embodiment of the present invention. FIG. 8 is a flowchart of a process for measuring the performance of a smart object.

FIG. 9 is a flowchart of a process of collecting the object arrangement information.

Figure 10 shows a display example of an object deployment diagram (flow type).

Figure 11 is a display example of an object deployment diagram (tree type). BEST MODE FOR CARRYING OUT THE INVENTION

· Assumed environment In the embodiment of the present invention, a system constituted by a plurality of hardware and software (applications) connected via a network, that is, hardware resources (computers such as servers and clients, networks, and the like) It is assumed that the system environment consists of a CPU, memory, hard disk, network card, etc.) and software resources (applications that are distributed and executed on hardware resources). It is assumed that some or all of the application will run under a virtual machine environment (CLR on .NET platform, JVM on J2EE platform, etc.).

• Virtual machine environment

Describe the virtual machine environment. In the apparatus of the present embodiment, when measuring the operation performance of the system (execution time of a processing unit, hardware processing performance, network communication time, and the like), measurement is performed via a virtual machine (VM). It also uses the VM when restoring source code from binary files. Environments where VMs can be used include Microsoft's .NET and J2EE proposed by Sun Microsystems. In this specification, a method (flow chart) based on a .NET environment will be described. Similar methods can be easily devised and implemented in other VM environments based on the methods described in this specification.

Figure 1 shows the structure of the .NET framework. The inside of the .NET framework consists of three main components. CLR (Common Language Runtime) at the bottom is an engine (VM) for executing application / component. And there are classes and libraries on top of the application system. ASP .NET uses Web Services and Web applications, excluding Windows applications Class library to implement. Applications on the .NET framework are coded using the class libraries provided by the .NET framework. The generated source code is converted into executable code of the application or component by the compiler, and the generated code is not native code depending on specific CPU instructions, but managed code and managed code. This is the intermediate code called. When the application runs, the managed code is converted to the final native code by the CLR JIT compiler and executed.

The CLR requires managed code to have two main pieces of information:

U) MSIL (Microsoft Intermediate Language (middle j): Executable code (2) Metadata: Information about MSIL

(Information required by the CLR at the time of execution, such as internal methods / properties and external methods / properties used)

• Overview of the embodiment

In the embodiment of the present invention, the software of the system whose performance is to be measured is analyzed, the processing units (objects) constituting the software are extracted, and the relationship between the processing units is clarified and shown in the figure. This figure is called the call graph.) Restore source code from binary file (executable file) in generating call graph.

Next, the operation performance of each processing unit (time required for execution, communication time between each processing unit, etc.) is measured. This is achieved without altering the source code of the application by measuring the operation performance using data that can be read from the virtual machine. Then, it clarifies on which hardware resource each processing unit is operating among the hardware resources that make up the system. It also measures and displays the load status of each hardware resource at the execution stage of the process. · Detailed description of the embodiment

1. Restore source code

Even if the source code is not directly available, in the virtual machine environment assumed by the present invention, the source code is restored from the binary file by using the method of the present invention.

For example, in a .NET environment, a compiler converts source code written in a high-level language into a binary file. Figure 2 is a schematic diagram showing the process of generating a binary file from the source code and the process of restoring the source code from the binary file. 203 binary files are generated by compiling 202 from source code written in a high-level language such as 201 C #, C ++, VB (visual basic). When restoring the source code from a binary file, the source code is restored by decompiling .204.

FIG. 3 is a flowchart showing processing steps for restoring the source code. In step S301, all server machines in the system are searched to find a server on which the .NET environment is installed, and connect to it. In step S302, the binary files of the application on .NET of the connected server are collected. In step S303, the binary file is read, and in step S304, the metadata and MSIL (intermediate code) are extracted from the binary file using the MSIL disassembler engine. In step S305, syntax analysis of the MSIL of the module is performed, and in step S306, semantic analysis is performed. Scrutinize each keyword and its parameters and match it with the corresponding keyword in the original high-level language. During this process, Identify classes, methods, properties, etc. As a result, the source code is restored as in step S306.

2. Analysis of application

As a method of analyzing an application and generating a call graph, a method named the lazy worker method is used here. The Rage-Worker method is one of the methods to analyze the source code of the application in the system, extract the processing units that make up the application, and analyze the calling relationship between the processing units. Here, the processing unit refers to a detailed execution unit of an application such as a class, a method, a property, and an instance or an object generated from the class.

The Rage-Worker method consists of two stages: a preparation process and a tree generation process. In the preparation process, the “caller and callee” of the class are analyzed and acquired based on the source code. In the next tree generation process, based on the Caller and Callee information obtained in the preparation process, the call relationship of all the processing units that the processing unit directly or indirectly calls for any processing unit is a one-type. Generate as a data structure. The tree structure generated by the tree generation process is represented by “nodes” and “pointers”. A node corresponds to a processing unit, and a pointer corresponds to a call relation between processing units. The call graph is generated by drawing the generated library structure with an appropriate drawing application. In the Rage-Worker method, the source code is analyzed in the preparation process, and the tree generation process generates the tree structure without referring to the source code. As a result, a call graph starting from an arbitrary processing unit can be efficiently generated and drawn without re-analyzing the source code.

(1) Preparation process FIG. 4 is a flowchart showing an example of the processing steps of the preparation process of the Rage-Worker method.

'In the preparation process, first, a class list is prepared in step S401. The class list is a data structure for registering classes. Each class in the list has a tree structure, and stores information on the methods in each class and the methods of the caller and callee of the class. After setting the initial value of the class list to null (empty), the source file processing of step S402 is performed on all source codes in the application.

In the source file processing, the class is extracted by detecting the start of the description of the class in step S411 for all the source codes in the source file, and the class processing in step S412 is performed on the detected class.

In the class processing, the detected class is set as the own class in step S421, and the method or property is extracted by detecting the start of the description of the method or property in step S422 from all the descriptions in the class, and the detected method or property is extracted. The cola-core processing of step S423 is performed on the patty. In the cola-core processing, first, the method or property detected in step S431 is set as the own method, and in step S432, list / data processing described below is performed on the own class and the own method.

From all the descriptions in the own method, a description calling another method or property is detected in step S433, the class to which the method or property belongs is set as a call destination class in step S434, and step S435 Alternatively, the port property is used as the callee method. In step S436, list data processing is performed for the called class and the called method as well as in step S432. Then, in step S437, the callee method is added to the callee information of the own class,

In S438, the own method is added to the caller information of the callee class. In the list 'data processing, if the class does not exist, it is added in step S441. If the method does not exist in the class, it is added in step S443.

(2) Tree generation process

Analyzed in the preparation process · Generates a tree-type data structure for the specified class based on the obtained caller (Caller) · callee (Callee) information according to the processing steps shown in Fig. 5.

In the tree generation process, it is referred to whether the class specified in step S501 exists in the class list. If the class exists, a node corresponding to the class is generated in step S502, and the node Perform processing. If not, an appropriate error message is displayed in step S504, and the process ends. In the node 'pointer process, the current node is set as the current node in step S511, the call destination information of the current node is referred to in step S512, and if it is not null, all of the information registered as the information are searched in step S513. Generates the called node. This node is called a target node. In step S514, pointers from the current node to all target nodes are generated. In step S515, the same node-vointer processing is performed on all the generated target nodes.

As described above, the calling relation between the processing units can be generated and drawn at a detailed level.

3. System configuration

FIG. 6 is a diagram showing a configuration of a system in which a program according to the embodiment of the present invention is mounted. Reference numeral 601 denotes a monitoring dedicated terminal, and 602 denotes a monitoring target terminal. These terminals are connected via 603 LAN, WAN, Internet and other networks. 4. Software configuration

FIG. 7 is a conceptual diagram of a functional configuration of a program according to the embodiment of the present invention. Agent Handler 701, each server 1 ~! ! Is the software that supervises and manages Agents # 1 to #n. The agent handler 701 operates as a monitoring-only terminal that monitors each monitoring target terminal. Agents are software that runs on each server. Agents #l to #n are servers 1 to which the agent is running! ! Monitor the operating performance of Agent Handler Obtains performance information from agents via agents # 1 to #n running on servers l to n and consoles # 1 to # η to monitor performance of each server. I do. Agents # 1 to # η of servers 1 to η acquire OS performance information acquisition means, application performance information acquisition means, database system (DBMS) performance information acquisition means, and network performance information, respectively. Means.

OS performance information is obtained from the OS level monitoring program. The OS level monitoring flowchart is, for example, WMI (Windows Management Instrumentation) in the Windows environment. Application performance information is obtained using the management service program. The management service program is a utility such as CLR (Common Language Runtime: having the function of a virtual machine) provided in the Microsoft (registered trademark) .NET platform. In the present invention, the performance information of the application is measured at a more detailed level by measuring the operation performance of the processing unit described later. Database performance information can be obtained by accessing the database directly. When the system shown in Fig. 6 is configured with a TCP / IP network, network performance information can be obtained by monitoring network devices using SNMP and monitoring the connection status of the line. 5. Measurement of operation performance for each processing unit

When measuring the operation performance (execution time, etc.) of a processing unit, in the virtual machine environment presupposed by the present invention, the operation performance is measured without any modification to the source code by the method according to the present invention.

Under the VM environment, the VM manages code execution as an application execution engine and provides various services to the application. At the time of application execution, the VM performs various managements related to the execution of each processing unit. Therefore, in the present invention, by introducing small software called a hooker into the VM, it is possible to freely communicate directly with the VM and acquire data relating to the measurement of operation performance from the VM. When executing an application, the VM manages the start, end, and call of the processing unit. When these events occur in the VM, the hooker detects the event and collects necessary information to measure the operation performance of the processing unit.

FIG. 8 is a flowchart showing processing steps for measuring the operation performance of each processing unit. This process starts when the fulfilment detects the above-mentioned VM event. -In S801, access the VM and check the type of event.

If the event type is generation of a processing unit in step S802,

In step S803, the metadata is accessed, and information necessary for performance measurement such as the class name, the method / property included in the inside, the external class to be used, and the method / property to be used is collected. Save time.

If the event type is a call of a processing unit in step S805, it is determined in step S806 whether the call destination is a processing unit executed by another server. In step S807, the time required for communication between the servers is recorded. If the type of event is the end of the processing unit in step S808, the total execution time is calculated in step S809, and stored in step S810. 6. Hardware resource identification means

In a VM environment, the information of application 'components placed on the server can be collected by referring to the metadata. This specifies the processing unit that is placed on the server and executed.

Figure 9 is a flow for collecting information on the arrangement status of processing units. In step S901, a connection is made to a server connected to the network. In step S902, a connection is made to the VM of the server, and in step S903, the metadata is referred to, and in step S904, information of a processing unit that can be executed in the server is collected. Save the information collected in step S905. By executing the above processing for all servers to be monitored, it is possible to specify the server on which each processing unit is executed.

7. System performance display

The operation performance of the processing unit measured by the above means can be expressed as an object deployment diagram (hereinafter referred to as ODD) together with the calling relationship between the processing units, the operation performance of the processing unit, and the hardware resources used by the processing unit. It is displayed according to the display method named "abbreviation").

Fig. 10 and Fig. 11 show ODD display examples. Cold Daraf (Graph showing the relationship between processing units) occupies most of the display area. Power is shown in Fig. 10 and in tree form in Fig. 11.

At the top of the figure, there are the display units of “Web server”, “Application server # 1” and “Application server # 2”. Each display unit has a color and indicates a server machine that constitutes a live application system. If Application # 2 is shown in red, and "Billing Process I was shown in red" Then, you can see that "billing process" is running on "application server # 2". All machine names in the system are displayed in different colors. The color of the text displayed on the screen corresponds to the color of the server machine.

On the right side of the figure is a graph labeled "CPUJ" "Memory J" "Disk" "Network". These graphs show the amount of hardware resources consumed by each server. This allows the user to check the current state of the hardware resources and get an association between the state of the hardware and a particular processing unit on the machine.

On the call graph, the average value of the operation performance at the time interval arbitrarily set by the user is shown. This operation performance includes network performance (communication time), operation performance per processing unit (execution time), and the like. The ODDs shown in the cases of Figs. 10 and 11 mainly deal with network performance and operating performance of processing units, and do not show elements that can be ignored in the operating performance of other systems.

There is a table at the bottom of the screen. This table is used to display detailed numerical information. Also, the user can obtain the history of specific data from the table. The display on this screen is updated at any time specified by the user. Users use this screen to identify problems when they want to see the performance of the system at a more detailed level, such as when a failure occurs that degrades the performance of the system. Failure factors can be categorized into any one of "application", "environment setting", "hardware resource", and "network", and correspond. "Hardware resources" refers to a failure of IT equipment such as a hard disk failure or display failure, and can be dealt with by repairing or replacing the failed part. In the case of "network", the range in which failures can be handled is limited to a specific range such as LAN. In addition, it is "environment setting" that sets these hardware resources ゃ network, but the hardware resources, that is, CPU, hard disk, Since applications such as memory and network are operated by applications, it can be said that most of system failures are caused by "applications". An example in which a failure is caused by each of "application", "environment setting", "hardware resource" and "network" will be described below.

1. For application

When an application is causing a failure, the system administrator can easily compare the “Average Response Time” and “Response Time” values in the table to easily determine that the performance of a particular processing unit has deteriorated. You can notice. If there is a large difference between these two values, we can assume that there is something wrong. In such a case, it is possible to examine the operation performance of other processing units and to specify the processing unit whose operation performance is most deteriorated. For example, if the current response time of a processing unit is 8.1 seconds and the average response time is 6.1 seconds, then it is possible that the system has failed. The system administrator also checks other running processing units. If the current response time of an "accounting" with an average response time of 3.1 seconds is 5.1 seconds, then the "accounting" has a difference of 2 seconds compared to the average. On the other hand, if the response time of other processing units is within 0.5 seconds, the system administrator can click "Accounting" and check the source code to determine the cause of the failure. .

2. For environment settings

If there is a problem with the environment setting, it affects the entire server machine, and the performance of the entire server machine will be degraded. A typical example is preferences The problem is that the operation performance of all processing units on the server machine deteriorates. The basic idea is similar to that of an application, but there are differences in the range that affects performance.

Faults can be found by carefully observing the degradation of operating performance, as in the case of applications.

3. For hardware resources

If there is a problem with the hardware resources, the consumption time of the CPU, memory, hard disk, etc. of a specific machine may not be obtained or the consumption rate may become 0, which is lower than the case of the application or environment setting. Thus, finding obstacles is easy.

For example, if the value of hardware resource consumption from Server A is no longer displayed in ODD, Server A can be considered to have failed.

4. For networks

Generally, network failures affect the server machines that make up the system when they are distributed over two or more LANs connected via the Internet. Even if the LAN of each site has good performance, it cannot guarantee the performance of the Internet.

For example, if server machine A is located on LAN # 1 and server machine B is located on LAN # 2, and LAN # 1 and LAN # 2 are connected via the Internet, how is LAN # 1 and LAN # 2 connected? Even at high speeds, no one can guarantee the network performance of the entire system because the Internet part cannot be controlled. In the ODD display, the colors and types of lines connecting individual objects (waves, dotted lines, etc.) are displayed differently to indicate that the objects exist on different networks. In addition, the performance of each display item is displayed.

According to the present invention, the operation performance of an application can be measured in finer processing units, and the correspondence with the hardware resources in which each processing unit is running can be easily clarified. In addition, ODD makes it possible to provide a visual and intuitive understanding of the correspondence between the performance of the individual processing units and the hardware resources on which they operate.

In this way, ODD makes it easy for users to investigate response performance by providing intuitively understandable information that helps identify the cause of a problem.

As a result, when system performance degradation is detected, it is possible to identify the cause at a fine processing unit level, and it is possible to know at a finer processing unit level the programs that need to be improved to solve them It becomes.

In addition, since it is possible to easily understand the correspondence between processing units and hardware resources, it is possible to improve the processing method of software and improve the processing capacity of IT resources when solving system performance degradation. The factors that need to be specified at a more detailed level in terms of both software and hardware that make up the system, such as planning, and items to be improved (such as processing units to be corrected or hardware resources to be enhanced) It becomes possible to know. As a result, it is possible to quickly optimize the operation performance of the application by rewriting or modifying the source code of the application and increasing the hardware resources.

Claims

The scope of the claims

1. Measure and analyze the performance of multiple applications installed on multiple servers connected via a network, some or all of which cooperate in a virtual machine environment. A program that enables a computer to perform system performance measurement and analysis to optimize cooperative operation based on the results.

An application analysis step of analyzing and extracting processing units constituting the application, and analyzing and acquiring a calling relationship between the processing units;

An operation performance measuring step of measuring the operation performance of the processing unit;

Analyzing the correspondence between the processing unit and the hardware resources used by the processing unit;

A display step of displaying together a calling relationship between the processing units, an operation performance of the processing units, and a hardware resource used by the processing units;

A system performance measurement analysis program characterized by comprising:

2. The system performance measurement analysis program according to claim 1, wherein the display of the display step is updated at an arbitrary time designated by a user, and indicates an operation performance of a processing unit at the time of the update.

3. The system performance measurement analysis according to claim 1, wherein in the operation performance measurement step, an execution time of a processing unit and a communication time of a network are measured.

4. In the operation performance measurement step, modify the application source code. 2. The system performance measurement analysis program according to claim 1, wherein the operation performance is measured without performing the operation.

5. In the application analysis step, the source code is restored using the disk assembler function provided by the virtual machine environment, and the source code is analyzed to extract the processing unit and obtain the calling relationship between the processing units. The system performance measurement analysis program according to claim 1, wherein the program is configured by the following steps.

6. The hardware resource identification step according to claim 1, wherein a correspondence between a processing unit and a hardware resource used by the processing unit is acquired using metadata of a virtual machine environment. System performance measurement analysis program.

7. The system performance measurement analysis program according to claim 1, wherein the display step visually represents a correspondence between a processing unit and a hardware resource on which the processing unit operates.

8. Measure and analyze the performance of multiple applications installed on multiple servers connected over a network, some or all of which cooperate in a virtual machine environment. A system performance measurement and analysis method for optimizing cooperative operation based on results.

An operation performance measurement method for measuring the operation performance of the processing unit. Analyzing the correspondence between the processing unit and the hardware resources used by the processing unit;

A method for measuring and analyzing system performance, comprising:

9. For multiple applications installed on multiple servers connected via a network, some or all of which cooperate in a partial machine environment, measure and analyze the performance of the applications and analyze the performance. A system performance measurement and analysis device for optimizing a cooperative operation based on a result.

Operating performance measuring means for measuring the operating performance of the processing unit;

Hardware resource specifying means for analyzing and acquiring the correspondence between the processing unit and the hardware resources used by the processing unit;

Display means for displaying together a call relationship between the processing units, an operation performance of the processing units, and a hardware resource used by the processing units;

A system performance measurement / analysis device comprising: