JP6093962B2 - Analysis system, computer system, and analysis method - Google Patents

Description

  The present invention relates to a technique for analyzing behavior, performance, and the like related to a database (DB).

  Currently, there are many applications based on DBs, and DBMSs that perform a series of processing and management related to DBs are extremely important. One of the features of DBMS is that it handles a large amount of data. Therefore, in many computer systems in which a DBMS operates, a system configuration in which a storage device having a large-capacity disk is connected to a computer in which the DBMS operates and the DB data is stored on the storage device is common.

  In the case of adopting this system form, data is stored on the disk of the storage device, so that access to the disk inevitably occurs when performing DB processing (DB processing). In particular, in a very large DB of petabyte class, it takes a lot of time to search for certain specific data from the DB data. Thus, a technique disclosed in Patent Document 1 is known as a technique for speeding up a search process for searching for specific data from a large amount of data.

  The technique disclosed in Patent Document 1 is a technique for dynamically creating a task each time data is read and multiplexing the data read by executing the tasks in parallel. According to a DBMS using this technology, search performance can be dramatically improved as compared with a conventional DBMS that executes tasks in the order of occurrence.

  The DBMS subdivides DB processing into processing units called tasks, executes tasks in high parallelism, issues I / O (asynchronous I / O) in a highly multiplexed manner, and maximizes storage performance. Improve performance. Therefore, the parallelism of processing (tasks) including I / O is very important for achieving performance. For example, it is important how many I / Os can be issued simultaneously to each HDD of the storage device, that is, how many tags (SCSI commands issued from the computer to the storage device) can be loaded on each HDD.

JP 2007-34414 A

  In a DBMS that executes processes (tasks) in parallel and issues I / O with high multiplex, the processing behavior becomes very complicated, and it can be used for operation debugging during development, performance debugging during system operation, etc. A huge amount of time is required for behavior and performance analysis.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide a technique capable of easily and appropriately analyzing the behavior, performance, and the like of a DBMS that executes tasks in parallel.

  In a DBMS that executes tasks in parallel, the processing parallelism, which is the number of tasks executed in parallel, is very important for achieving performance, and this processing parallelism is determined by the content of the query to be executed, hardware resources, etc. Is. In order to achieve the above object, the present invention focuses on this processing parallelism and facilitates the behavior and performance analysis of the DBMS.

  An analysis system according to an aspect of the present invention analyzes a behavior of a computer system in DB processing of a DBMS. The computer system has a storage device and a computer. The storage device has a plurality of storage media for storing DB data. The DBMS operates on a computer having a processor core, and can transmit a plurality of data read requests for executing a query to the DB to the storage device. The DBMS may dynamically generate a task for executing an operation and execute the dynamically generated task in query processing. Specifically, for example, in the query processing, the DBMS (a) generates a task for executing the operation, (b) executes the generated task, (c) in (b) When the (N + 1) th operation is executed based on the execution result of the Nth operation corresponding to the executed task, a task based on the execution result is newly generated (N is an integer of 1 or more); And (d) performing (b) and (c) on the newly generated task. In the above (b) and (d), when there are two or more executable tasks, the DBMS may execute at least two of the two or more tasks in parallel. . The task may be executed by using a thread (process) managed by the OS. A plurality of threads executed by the processor core may execute the task, and each thread may execute a plurality of tasks. The operation of the DBMS may be an operation according to the technique disclosed in Patent Document 1 described above.

  The analysis system includes a storage resource capable of storing information and a processor that executes processing.

  The storage resource can specify the number of threads specifying information that can specify the maximum number of threads for the query in the DBMS and the number of first input / output processes that can be executed in parallel at the interface with the storage device of the computer. 1 process number specifying information, second process number specifying information capable of specifying the number of second input / output processes that can be executed in parallel with respect to the storage medium in the storage device, and third input / output capable of executing each storage medium in parallel The third process number specifying information that can specify the process number is stored.

  The processor acquires the number of selected rows corresponding to a key value of an index key related to a query from the DBMS, and the number of selected rows, the maximum number of threads specified by the thread number specifying information, and the first processing number The first input / output process number specified by the specific information, the second input / output process number specified by the second process number specifying information, and the third input / output process number specified by the third process number specifying information Based on the above, a process parallelism model predicted value that is a process parallelism predicted based on a model in the process corresponding to the query is calculated. The processor acquires event information related to an input / output event for the storage medium when the processing corresponding to the query is actually executed from the storage device, and performs the processing corresponding to the query based on the event information. The process parallelism actual measurement value which is the process parallelism at the time of execution is calculated, and control based on the process parallelism model predicted value and the process parallelism actual measurement value is displayed.

  With the analysis system according to the present invention, it is possible to easily and appropriately analyze the behavior, performance, and the like of a DBMS that executes tasks in parallel.

FIG. 1 is a configuration diagram of an example of a computer system according to the first embodiment. FIG. 2A is a configuration diagram of an example of schema information according to the first embodiment. FIG. 2B is a configuration diagram of an example of DB file information according to the first embodiment. FIG. 2C is a configuration diagram of an example of DB statistical information according to the first embodiment. FIG. 2D is a configuration diagram of an example of query plan information according to the first embodiment. FIG. 2E is a configuration diagram of an example of DB processing information according to the first embodiment. FIG. 3A is a configuration diagram of an example of OS mapping information according to the first embodiment. FIG. 3B is a configuration diagram of an example of OS processing information according to the first embodiment. FIG. 4A is a configuration diagram of an example of ST mapping information according to the first embodiment. FIG. 4B is a configuration diagram of an example of ST processing information according to the first embodiment. FIG. 5A is a configuration diagram of an example of system information according to the first embodiment. FIG. 5B is a diagram illustrating an example of a query (SQL) according to the first embodiment. FIG. 5C is a diagram illustrating an example of a query plan according to the first embodiment. FIG. 6 is a flowchart of the analysis / visualization process according to the first embodiment. FIG. 7 is a first flowchart of the process parallelism model predicted value calculation process according to the first embodiment. FIG. 8 is a second flowchart of the process parallelism model predicted value calculation process according to the first embodiment. FIG. 9 is a flowchart of the process parallelism actual value calculation process according to the first embodiment. FIG. 10A is a diagram illustrating an example of a process parallelism model predicted value graph according to the first embodiment. FIG. 10B is a diagram illustrating a first example of a graph of the process parallelism model predicted value and the process parallelism actual measurement value according to the first embodiment. FIG. 10C is a diagram illustrating a second example of the graph of the process parallelism model predicted value and the process parallelism actual measurement value according to the first embodiment. FIG. 11A is a diagram illustrating a screen display example including a graph of the processing parallelism model predicted value according to the first embodiment. FIG. 11B is a diagram illustrating a screen display example including a graph of the process parallelism model predicted value and the process parallelism actual measurement value according to the first embodiment. FIG. 11C is a diagram illustrating a screen display example including a graph regarding the staying number according to the first embodiment. FIG. 12A is a diagram illustrating a screen display example including a graph indicating conformity / nonconformity between the model predicted value and the actual measurement value of the processing parallelism in the HDD according to the first embodiment. FIG. 12B is a diagram illustrating a screen display example indicating conformity / nonconformity between the model predicted value of the processing parallelism and the actual measurement value for each HDD according to the first embodiment. FIG. 12C is a diagram illustrating a screen display example in which HDDs having a high mismatch rate between the model predicted value and the actual measurement value of the processing parallelism according to the first embodiment are sequentially illustrated. FIG. 13A is a diagram illustrating a screen display example including a table accessed by the DBMS according to the first embodiment and a graph of a statistical value of the number of tasks for each index. FIG. 13B is a diagram illustrating a first screen display example including a graph of statistical values for each task state according to the first embodiment. FIG. 13C is a diagram illustrating a second screen display example including a graph of the state of each task according to the first embodiment. FIG. 14 is a diagram illustrating a third example of the graph of the process parallelism model predicted value and the process parallelism actual measurement value according to the first embodiment. FIG. 15A is a configuration diagram of an example of logical volume information according to the first embodiment. FIG. 15B is a configuration diagram of an example of volume group information according to the first embodiment. FIG. 16 illustrates an example of a schema group according to the second embodiment. FIG. 17A illustrates an example of a query according to the second embodiment. FIG. 17B illustrates an example of a query plan according to the second embodiment. FIG. 17C illustrates an example of a query result according to the second embodiment. FIG. 18 shows the data structure of the index IA and the relation table A according to the second embodiment. FIG. 19 shows the data structure of the index IB and the relational table B according to the second embodiment. FIG. 20 illustrates an example of a data structure summary table according to the second embodiment. FIG. 21 shows that a conventional DBMS executes tasks sequentially. FIG. 22 shows an example of DB processing information of a conventional DBMS. FIG. 23 illustrates that the DBMS according to the second embodiment executes tasks in parallel. FIG. 24 illustrates an example of DB processing information of the DBMS according to the second embodiment. FIG. 25 illustrates a visualization window according to the second embodiment. FIG. 26 illustrates an example of a task behavior visualization tree according to the second embodiment. FIG. 27 illustrates an example of a data structure behavior visualization tree according to the second embodiment. FIG. 28 shows an example of a task behavior visualization tree after the designated time is changed. FIG. 29 shows an example of the data structure visualization tree after the designated time is changed. FIG. 30 illustrates an example of a radical tree as a modification of the task behavior visualization tree according to the second embodiment. FIG. 31 shows an example of a radical tree after the designated time has been changed.

  Several embodiments of the present invention will be described with reference to the drawings. The embodiments described below do not limit the invention according to the claims, and all the elements and combinations described in the embodiments are essential for the solution of the invention. Is not limited.

  In the following description, processing may be described using “program” as the subject, but the program is defined by being executed by a processor (for example, a CPU (Central Processing Unit)) included in a computer, a storage device, or the like. Therefore, the subject of the processing may be a processor, using a storage resource (for example, a memory) and / or a communication interface device (for example, a communication port) as appropriate. The processing described with the program as the subject may be processing performed by a processor or a device (computer, storage device, etc.) having the processor. The controller may be the processor itself or may include a hardware circuit that performs part or all of the processing performed by the controller. The program may be installed in each controller from a program source. The program source may be, for example, a program distribution computer or a storage medium.

  In addition, a serial interface or an Ethernet interface (Ethernet is a registered trademark) is used as an input / output device as an alternative to an input / output device in a computer, and a display device having a display, a keyboard, or a pointer device is connected to the interface. May be sent to the display device or input information may be received from the display device to display on the display device or accept input to replace the input and display on the input / output device. .

  Hereinafter, a set of one or more computers that analyze various types of information in the computer system may be referred to as an analysis system. When the computer displays the display information, the computer is an analysis system. A combination of a computer and a display device is also an analysis system. Further, analysis and display processing may be realized by a plurality of computers in order to increase the speed and reliability of the analysis processing. In this case, the plurality of computers (in the case where the display device performs the display, the display device) Analysis system).

  FIG. 1 is a configuration diagram of an example of a computer system according to the first embodiment.

  The computer system includes a computer 100, a storage device 150 as an example of a storage device, and a computer 180 as an example of an analysis system. In FIG. 1, only one computer 100 and one storage device 150 are shown, but a plurality of them may be used. The computer 100 and the storage device 150 are connected via a communication network 140. The storage device 150 stores DB data. The computer 100 manages DB data stored in the storage device 150. Hereinafter, a person who operates the computer 100 is referred to as a “system administrator” for convenience, but the computer 100 may be operated by a person other than the system administrator.

  The computer 100, the storage device 150, and the computer 180 are connected via a communication network 142. The computer 180 acquires various types of monitor information from the computer 100 and the storage device 150 and performs analysis / visualization processing. The communication networks 140 and 142 may be a network such as a local area network (LAN) or a wide area network (WAN), or may be a network (storage area network: SAN) configured with a fiber channel or the like.

  The computer 100 can be realized by a general computer, for example. For example, the computer 100 includes a CPU (control processor) 104, an input / output device 106, a storage device 108, a memory 110, an I / F (1) 136, and an I / F (2) 138. The CPU 104, the input / output device 106, the storage device 108, the memory 110, the I / F (1) 136, and the I / F (2) 138 are connected via the internal bus 102.

  The I / F (1) 136 is an interface with the communication network 140, and the I / F (2) 138 is an interface with the communication network 142. The I / F (1) 136 is, for example, an HBA (Host Bus Adapter). The HBA has one or more ports. The input / output device 106 includes, for example, an input device such as a mouse and a keyboard, and an output device for a liquid crystal display. The storage device 108 stores a program executed by the CPU 104, information required by the CPU 104, and the like.

  The memory 110 stores a program executed by the CPU 104, information required by the CPU 104, and the like. For example, the memory 110 stores an operating system (hereinafter referred to as OS) 112 and a DBMS 120. The OS 112 associates a device managed by the OS 112 with a logical storage area on the storage apparatus 150, and OS processing information 116 regarding processing (processing event) in the OS when the DBMS 120 executes a query. And hold. The OS 112 also includes a program for configuring the OS monitor information output unit 118 that outputs the OS mapping information 114 and the OS processing information 116 to the outside. The OS monitor information output unit 118 is configured by the CPU 104 executing the program. When the monitor information is output to the computer 180, the OS monitor information output unit 118 may once write it out to a file and then transfer it to the computer 180 or directly transfer it to the computer 180.

  The DBMS 120 includes schema information 122 relating to schemas (hereinafter also referred to as objects) of DB tables and indexes, DB file information 124 relating to files storing DB data, DB statistical information 126 relating to statistics inside the DBMS 120, and DBMS 120 The query plan information 128 related to the query to be executed and the DB processing information 130 related to the processing when the DBMS 120 executes the query are stored. The DBMS 120 also includes a program for configuring the DB monitor information output unit 132 that outputs the above information to the outside. The DB monitor information output unit 132 is configured by the CPU 104 executing the program. When the monitor information is output to the computer 180, the DB monitor information output unit 132 may once write the file to a file and then transfer it to the computer 180 or directly transfer it to the computer 180.

  The CPU 104 executes various processes by executing programs stored in the memory 110. For example, the CPU 104 executes the DBMS 120 to accept a query from an application (not shown), creates a query plan for the accepted query, and executes processing according to the query plan. In addition, when it is necessary to access DB data stored in the storage apparatus 150 during this processing, the CPU 104 sends an access request (I / O request) for the corresponding data via the OS 112 to the storage apparatus. Issue to 150.

  1 illustrates an example in which both the OS 112 and the DBMS 120 are stored in the memory 110, but at least a part of the OS 112 and the DBMS 120 may be stored in the storage device.

  The storage device 150 includes a controller 154 and a plurality of disks (HDD) 178 as storage media. The controller 154 and the plurality of disks 178 are connected by an internal bus 152. The disk 178 is, for example, a hard disk drive (HDD, magnetic storage device). In the storage apparatus 150, a plurality of disks 178 may be configured as a RAID (Redundant Array of Independent (or Inexpensive) Disks). In addition to the disk 178 or instead of the disk 178, the storage apparatus 150 may include a storage device (for example, a flash memory drive) having other types of storage media.

  The controller 154 includes an I / F (1) 158, an I / F (2) 160, a CPU (control processor) 162, a cache memory 164, and a memory 168. The I / F (1) 158, the I / F (2) 160, the CPU 162, the cache memory 164, and the memory 168 are connected by, for example, an internal bus 156. The I / F (1) 158 is an interface with the communication network 140, and the I / F (2) 160 is an interface with the communication network 142. The cache memory 164 stores necessary data.

  The memory 168 stores programs executed by the CPU 162, information required by the CPU 162, and the like. The memory 168 stores a control program 170 that controls the storage apparatus 150. The control program 170 includes ST mapping information 172 that associates a logical storage area (LU: Logical Unit) of the storage apparatus 150 with a physical storage area of the disk 178, and a storage apparatus when the DBMS 120 executes a query. The ST processing information 174 related to the processing (processing event) in 150 is held. The control program 170 includes a program for configuring the ST monitor information output unit 176 that outputs the above information to the outside. The ST monitor information output unit 176 is configured by the CPU 162 executing the program. When transmitting the ST processing information 174 to the computer 180, the ST monitor information output unit 176 may once write the file to a file and transmit it to the computer 180 collectively after the query ends, or an entry is added to the ST processing information 174. It may be transmitted directly to the computer 180 each time. The CPU 162 executes various processes by executing a program stored in the memory 168.

  The computer 180 can be realized by, for example, a general computer. For example, the computer 180 includes an I / F 184, a CPU (control processor) 186, an input / output device 188, a storage device 190, and a memory 192. The I / F 184, the CPU 186, the input / output device 188, the storage device 190, and the memory 192 are connected by an internal bus 182. Here, the storage device 190 and the memory 192 are examples of storage resources.

  The I / F 184 is an interface with the communication network 142. For example, an input device such as a mouse and a keyboard and a display 144 are connected to the input / output device 188. The display 144 displays various information, for example, execution results by the analysis / visualization program 195. The storage device 190 stores programs executed by the CPU 186, information required by the CPU 186, and the like.

  The memory 192 stores programs executed by the CPU 186, information required by the CPU 186, and the like. The memory 192 includes system information 196 that is information related to the computer system, DB monitor information 197 that is a series of information output by the DB monitor information output unit 132, and an OS monitor that is a series of information output by the OS monitor information output unit 118. Information 198, ST monitor information 199 which is a series of information output by the ST monitor information output unit 176, and an analysis / visualization program 195 are stored. The system information 196 may be given as a parameter by the system administrator, or may be automatically acquired from the DB monitor information output unit 132, the OS monitor information output unit 118, and the ST monitor information output unit 176. good. The CPU 186 executes various processes by executing programs stored in the memory 192. For example, the CPU 186 executes processing using the system information 196 and various monitor information stored in the memory 192 by executing the analysis / visualization program 195.

  1 shows an example in which both the OS 194 and the analysis / visualization program 195 are stored in the memory 192. However, at least a part of the OS 194 and the analysis / visualization program 195 is stored in the storage device 190. May be.

  Next, details of the various information described above will be described.

  FIG. 2A is a configuration diagram of an example of schema information stored in the DBMS according to the first embodiment.

  The schema information 122 is information related to objects such as tables and indexes constituting the DB, and has an entry for each object. Each entry includes a field 200 for registering an identifier (Object ID) for identifying an object, a field 202 for registering an object name, a field 204 for registering an object type (table, index), and a data amount of the object. , A field 208 for registering the number of object rows (only when the object type is a table), and a field 209 for registering an identifier (File-ID) of a DB file storing the object data. The schema information 122 is created when a DB is constructed, and updated when an object is added or deleted.

  FIG. 2B is a configuration diagram of an example of DB file information stored in the DBMS according to the first embodiment.

  The DB file information 124 is information related to a DB file in which DB data is stored, and has an entry for each DB file. Each entry has a field 210 for registering an identifier (File-ID) for identifying a DB file, and a field 212 for registering a device name on the OS 112 where the DB file is created. The DB file information 124 is created when a DB is constructed, and updated when a DB file is added / deleted / changed.

  FIG. 2C is a configuration diagram of an example of DB statistical information stored in the DBMS according to the first embodiment.

  The DB statistical information 126 is information related to the internal statistics of the DBMS 120, and registers a field 220 for registering sequential I / O performance, and random I / O performance as an example of input / output time information indicating time related to random input / output. The field 222 includes fields 224 and 226 for registering the number of selected rows for the key value of the index. The number of selected rows 224 for the key value of the index has an entry for each key value of the index. The DB statistical information 126 is appropriately updated when the DBMS 120 executes DB processing. The information regarding the number of selected rows for the index key value may be information calculated based on information such as a histogram, or may be a parameter directly given from the outside. Further, the values registered in the sequential I / O performance 220 and the random I / O performance 222 may be the average values of the past past I / O response times in a certain period.

  FIG. 2D is a configuration diagram of an example of query plan information stored in the DBMS according to the first embodiment.

  The query plan information 128 is information related to a query executed by the DBMS 120, and has an entry for each scan in the query. Each entry includes a field 230 for registering an identifier (QID) for identifying a query, a field 232 for registering an identifier (SubQID) for identifying a processing block (partial query) in the query, and a scan in the query. A field 234 for registering an identifier (ScanID) for registration, a field 236 for registering the type of scan in the query, a field 238 for registering an identifier for identifying an object to be accessed in the scan, and the scan is an index scan (Index Scan) In this case, a field 240 for storing an index key serving as a search condition and a field 242 for registering an index key value serving as a search condition are provided. The query plan information 128 is created after the DBMS 120 accepts the query. Based on the information, the query plan information 128 indicates which table / index the target query accesses in what order (index key, key value) and in what order. Can be identified.

  FIG. 2E is a configuration diagram of an example of DB processing information stored in the DBMS according to the first embodiment.

  The DB processing information 130 is information related to processing when the DBMS 120 executes a query, and an entry (event) when a specific event (for example, asynchronous I / O issue event, asynchronous I / O reaping event) is executed in the processing. Information). Each entry has a field 250 for registering an entry number, a field 252 for registering a time stamp of the time when the event was executed, a field 254 for registering an identifier (Thread ID) for identifying the thread that executed the event, and an event type. A field 256 to be registered, a field 258 for registering an identifier (Task ID) for identifying a task (task in a thread) that executed an event, and a task that generates a task identified from the identifier registered in the field 258 (parent) A field 260 for registering an identifier (Present Task ID) for identifying a task), a field 262 for registering an identifier (QID) for identifying an event execution source, and a query in the query for event execution Register an identifier (SubQID) to identify the processing block Field 264 for registering an identifier (Object ID) for identifying an access destination object, a field 268 for registering an identifier (File-ID) for identifying an access destination DB file, and the data to be accessed It has a field 270 for registering an offset, a field 272 for registering the size of data to be accessed, and a field 274 for registering an identifier (IO-ID) for identifying the I / O issued by the event.

  FIG. 3A is a configuration diagram of an example of OS mapping information according to the first embodiment.

  The OS mapping information 114 is information that associates a device managed by the OS 112 with a logical storage area (LU) on the storage apparatus 150, and has an entry for each device. Each entry includes a field 300 for registering a device name for identifying a device, a field 302 for registering an identifier (ST-ID) for identifying a storage apparatus 150 having a storage area (LU) corresponding to the device, It has a field 304 for registering a number (LUN) for identifying a storage area (LU) corresponding to the device. The OS mapping information 114 is created when the system is constructed, and is updated when the system configuration is changed.

  FIG. 3B is a configuration diagram of an example of OS processing information according to the first embodiment.

  The OS processing information 116 is information relating to processing in the OS when the DBMS 120 executes a query, and executes a specific event (for example, asynchronous I / O acceptance, I / O request issuance, I / O completion acceptance, etc.). Entry (event information) is created. Each entry includes a field 310 for registering an entry number, a field 312 for registering a time stamp of the time when the event was executed, a field 314 for registering an event type, and an identifier (IO for identifying an I / O corresponding to the event -ID) field 316, an identifier (ST-ID) for registering the access destination storage apparatus 150, a field 318 for registering, and an access destination logical storage area (LU) number ( LUN), a field 322 for registering a logical address (LBA) of data to be accessed, and a field 324 for registering the size of the data to be accessed.

  FIG. 4A is a configuration diagram of an example of ST mapping information according to the first embodiment.

  The ST mapping information 172 is information that associates a logical storage area (LU) managed by the storage apparatus 150 with a physical storage area of the disk 178, and has an entry for each LU. Each entry constitutes a field 400 for registering an identifier (ST-ID) for identifying the storage apparatus 150 having the LU, a field 402 for registering a number (LUN) for identifying the LU, and the LU. It has a field 404 for registering the number of disks 178 and a field 406 for registering an identifier (HDD-ID) for identifying the disk 178 constituting the LU concerned. The ST mapping information 172 is created when the system is constructed, and is updated when the system configuration is changed. When one LU is composed of a plurality of disks 178, a plurality of fields 406 are created in the same entry.

  FIG. 4B is a configuration diagram of an example of ST processing information according to the first embodiment.

  The ST processing information 174 is information related to processing in the storage apparatus 150 when the DBMS 120 executes a query, and specific events (for example, I / O request acceptance, disk I / O start, disk I / O end, I An entry (event information) is created when an / O response issuance is executed. Each entry has a field 410 for registering an entry number, a field 412 for registering a time stamp of the time when the event was executed, a field 414 for registering an event type, and an identifier (IO) for identifying an I / O corresponding to the event. -ID) field 416, I / O request size field 418, access destination logical storage area (LU) identification number field (LUN) field 420, access data A field 422 for registering the logical address (LBA) of the data, a field 424 for registering an identifier (HDD-ID) for identifying the disk 178 storing the data to be accessed, and the physical address of the data to be accessed ( PBA) is registered.

  FIG. 5A is a configuration diagram of an example of system information according to the first embodiment.

  The system information 196 is information related to the computer system, and includes a field 500 for registering the number of kernel threads for executing DB processing, a field 502 for registering the number of tasks per kernel thread for executing DB processing, and an HBA (Host Bus Adapter). ) For registering the number of simultaneous I / O processes per HBA port, a field 508 for registering the bandwidth of the communication network 140, and a field 510 for registering the number of controllers 154 of the storage apparatus 150. It has a field 512 for registering the number of simultaneous I / O processes per storage controller, a field 514 for registering the number of disks, and a field 516 for registering the number of simultaneous I / O processes (tags) per disk. Here, the number of kernel threads and the number of tasks per kernel thread correspond to the thread number identification information, and the number of HBA ports and the number of simultaneous I / O processes per HBA port correspond to the first process number identification information. The number of controllers 154 and the number of simultaneous I / O processes per storage controller correspond to the second process number specifying information, and the number of simultaneous I / O processes per one disk (number of tags) is the third process number specifying information. It corresponds to.

  Next, in describing the processing executed by the analysis / visualization program 195, an example of a query (SQL) for the DBMS 120 and a query plan for the query will be described.

  FIG. 5B is a diagram illustrating an example of a query (SQL) according to the first embodiment. FIG. 5C is a diagram illustrating an example of a query plan according to the first embodiment.

  In the query 520 shown in FIG. 5B, as shown in FIG. 5C, the portion “Select MAX (C2) From T2 Where C3 = 10” is first executed as the subquery 534 (processing block B: partial query), and then the subquery 520 is executed. Using the execution result of the query 534, the main query 532 (processing block A: partial query) of “Select C1 FromT1 Where C2 = (subquery result)” is executed by the join processing method using the index. Become.

  It can be seen that the processing corresponding to the query 520 is composed of two processing blocks: a processing block B corresponding to the sub-query 534 and a processing block A corresponding to the main query 532. The CPU 104 that executes the DBMS 120 identifies a query plan based on the query and registers the query plan in the query plan information 128.

  Next, processing by the computer 180 that executes the analysis / visualization program 195 will be described. Prior to the execution of the processing, it is assumed that the computer system construction and the DB construction have been completed, and the DBMS 120 of the computer 100 has accepted the query 520 and executed various processes. Therefore, schema information 122, DB file information 124, DB statistical information 126, DB processing information 128, OS mapping information 114, OS processing information 116, ST mapping information 172, ST processing information 174, and system information 196 are created. , The information at that time is reflected.

  FIG. 6 is a flowchart of the analysis / visualization process according to the first embodiment. The analysis / visualization process is a process realized by the CPU 186 executing the analysis / visualization program 195. Further, the processing parallelism, which is the number of tasks executed in parallel, is determined by the contents of the query to be executed, hardware resources, and the like. In the first embodiment, the final processing parallelism of the disk 178 of the storage apparatus 150 is The number of tags (hereinafter referred to as the number of HDD tags) is assumed.

  When the system administrator gives an instruction to start the analysis / visualization program 195, the computer 180 starts analysis / visualization processing (step 600). The processing in the following steps may be started automatically when the analysis / visualization program 195 is started, or may be started by an instruction from the system administrator.

  The analysis / visualization program 195 uses the monitor information (schema information 122, DB file information 124, DB statistical information 126, query plan information 128, and DB processing information 130) output by the DB monitor information output unit 132 as DB monitor information 197. Obtained and read into the memory 192 (step 602). Next, the monitor information (OS mapping information 114, OS processing information 116) output by the OS monitor information output unit 118 is acquired as OS monitor information 198 and read into the memory 192 (step 604). Next, the monitor information (ST mapping information 172, ST processing information 174) output by the ST monitor information output unit 176 is acquired as ST monitor information 199 and read into the memory 192 (step 606). Next, the system information 196 is read from the storage device 190 onto the memory 192 (step 608).

  Subsequently, the analysis / visualization program 195 executes the process parallelism model predicted value calculation process 700, and when the query 520 is executed in the computer system, the process / process is calculated as a numerical value indicating the process behavior predicted based on the model. A parallelism (HDD tag number) model prediction value is calculated (step 610). Next, the process parallelism actual value calculation process 800 is executed, and the process parallelism (HDD tag number) actual value is calculated as a value indicating the actual processing behavior when the query 520 is executed in the computer system (step 612).

  Subsequently, the analysis / visualization program 195 causes the display 144 to display a graph based on the process parallelism model predicted value calculated in step 610 and the process parallelism actual value calculated in step 612 (step 614). The process ends according to the instruction from the system administrator (step 616). In the first embodiment, on the display 144, a graph indicating the processing parallelism model predicted value and the processing parallelism actual measurement value on the vertical axis and the time on the horizontal axis is displayed. According to this graph, it is possible to easily compare the processing parallelism model predicted value and the processing parallelism actual measurement value. A detailed graph will be described later.

  FIG. 7 is a first flowchart of the process parallelism model predicted value calculation process according to the first embodiment. FIG. 8 is a second flowchart of the process parallelism model predicted value calculation process according to the first embodiment.

  The analysis / visible program 195 refers to the query plan information 128 in the DB monitor information 197 and specifies the number of processing blocks of the target query (step 702). For example, in the case of the query 520, the number of processing blocks is 2. Next, step 708 to step 742 are executed for each identified processing block. It is determined whether or not the processing parallelism model prediction values have been calculated for all the processing blocks (step 704), and when the processing parallelism model prediction values for all the processing blocks have been calculated (Yes in step 704), The process parallelism model predicted value calculation process is terminated (step 706).

  The analysis / visualization program 195 identifies the scan type of the processing block with reference to the query plan information 128 in the DB monitor information 197 (step 708) and determines whether the scan type is a table scan (Table Scan) or an index scan (Index Scan) (step 710). The table scan is a scan for acquiring records that meet the conditions by referring to only the table, and the index scan is a scan for acquiring records of the tables that match the conditions using the index. As a result of the determination, when the scan type of the processing block is a table scan (No in Step 710), 1 is set as the processing parallelism (Step 712). The table scan can also be processed in parallel, but the parallelism is much smaller than the processing parallelism assumed by the present invention, and is set to 1 here. On the other hand, when the scan type of the processing block is an index scan (Yes in step 710), the DB statistical information 126 in the DB monitor information 197 is referred to and corresponds to the index key and key value of the scan of the processing block. The number of key value selection rows is acquired, and the acquired value is set as the processing parallelism (step 714).

  For example, when processing block B of query 520 is targeted, the scan type of processing block B is index scan, and the index key is I2. C3, it can be identified from the query plan information 128 that the key value is C3 = 10, and I2. It can be identified from the DB statistical information 126 that the key value selection row number of C3 = 10 is 100,000. Therefore, the processing parallelism of processing block B at the end of step 714 is set to 100,000. When processing block A of query 520 is targeted, the scan type of processing block A is index scan, and the index key is I1. C2, it can be specified from the query plan information 128 that the key value is C2 = Max (C2), and I1. It can be identified from the DB statistical information 126 that the key value selection row number of C2 = Max (C2) is 100. Therefore, the processing parallelism of processing block A at the end of step 714 is set to 100. In this example, one processing block is equal to one scan, but a plurality of scans may be included in one processing block as long as the processing parallelism does not change.

  Subsequently, the analysis / visualization program 195 refers to the system information 196 and multiplies the number of kernel threads and the number of tasks per kernel thread to calculate the maximum parallelism of the DBMS 120 (step 716). In the case of the system information 196 shown in FIG. 5A, since the number of kernel threads is 64 and the number of tasks per kernel thread is 1,000, the maximum parallelism of the DBMS 120 is 64,000. The maximum degree of parallelism calculated here is based on the premise that the number of I / Os that can be issued simultaneously per task is 1. In the case of simultaneously issuing a plurality of I / Os per task, the result obtained by multiplying the number of kernel threads, the number of tasks per number of kernel threads, and the number of simultaneous I / Os issued per task is the maximum degree of parallelism.

  Next, the analysis / visualization program 195 compares the processing parallelism at this time with the maximum parallelism of the DBMS 120 calculated in step 716 (step 718), and if the maximum parallelism of the DBMS 120 is lower (in step 718) Yes) sets the maximum parallelism of the DBMS 120 as the processing parallelism (step 720). For example, when the processing block B of the query 520 is the target, the processing parallelism at this time (the processing parallelism of the processing block B set in step 714) is 100,000, and the DBMS 120 calculated in step 716 is the maximum. Since the parallel degree is 64,000 and the maximum parallel degree of the DBMS 120 is smaller, 64,000 is set as the processing parallel degree of the processing block B. Further, when the processing block A of the query 520 is the target, the processing parallelism (processing parallelism set in step 714) at this time is 100, and the maximum parallelism 64,000 of the DBMS 120 is larger. The processing parallelism of the processing block A remains 100. Note that the processing parallelism for each processing block at this time can be obtained by multiplying the maximum number of threads of the DBMS 120 (a value obtained by multiplying the number of kernel threads and the number of tasks per kernel thread) by a constant given from the outside. good.

  Subsequently, the analysis / visualization program 195 refers to the system information 196 and multiplies the number of HBA ports by the number of simultaneous I / O processes per number of HBA ports to obtain the number of HBA simultaneous I / O processes (first input / output process). Number) is calculated (step 722). In the case of the system information 196 shown in FIG. 5A, the number of HBA ports is 6, and the number of simultaneous I / O processes per HBA port is 256. Therefore, the number of HBA simultaneous I / O processes is 1,536.

  Next, the processing parallelism at this time is compared with the HBA simultaneous I / O processing number calculated in step 722 (step 724), and when the HBA simultaneous I / O processing number is lower (Yes in step 724). Sets the number of HBA simultaneous I / O processes in the processing parallelism (step 726). When processing block B of query 520 is targeted, the processing parallelism of processing block B at this point is 64,000, and the number of HBA simultaneous I / O processes calculated in step 722 is 1,536. Since the number of HBA simultaneous I / O processes is smaller, 1,536 is set to the processing parallelism of processing block B. When processing block A of query 520 is targeted, the processing parallelism at this point is 100, and the number of HBA simultaneous I / O processes 1,536 is larger, so the processing parallelism of processing block A Remains 100.

  Subsequently, the analysis / visualization program 195 refers to the system information 196 and multiplies the number of controllers of the storage apparatus 150 by the number of simultaneous I / O processes per controller to obtain the number of controller simultaneous I / O processes (second input). The number of output processes is calculated (step 728). In the case of the system information 196 shown in FIG. 5A, the number of controllers of the storage apparatus 150 is 2, and the number of simultaneous I / O processes per controller is 1,000. , 000.

  Next, the analysis / visualization program 195 compares the processing parallelism at this time with the controller simultaneous I / O processing number calculated in step 728 (step 730), and the controller simultaneous I / O processing number is lower. In the case (Yes in Step 730), the controller simultaneous I / O processing number is set as the processing parallelism (Step 732). On the other hand, when the number of simultaneous controller I / O processes is the same or high (No in step 730), the processing parallelism remains unchanged.

  For example, if the processing block B of the query 520 is the target, the processing parallelism of the processing block B at this time is 1,536, and the controller simultaneous I / O processing number calculated in step 728 is 2,000. Since the number of controller simultaneous I / O processes is larger, the processing parallelism of the processing block B remains 1,536. When the processing block A of the query 520 is the target, the processing parallelism at this point is 100, and the parallelism of the processing block A is 100 because the controller simultaneous I / O processing number 2,000 is larger. Will remain.

  Subsequently, the analysis / visualization program 195 refers to the schema information 122 and DB file information 124 of the DB monitor information 197, the OS mapping information 114 of the OS monitor information 198, and the ST mapping information 172 of the ST monitor information 199, The number of HDDs of the disk 178 storing the data of the object accessed by the block is calculated (step 734), and the processing parallelism at this time is divided by the calculated number of HDDs to calculate the processing parallelism per HDD. (Step 736). When processing block B of query 520 is targeted, the object accessed in processing block B is “T2” (access destination Object ID 238 of query plan information 128), and the DB file storing “T2” is “FILE2” (File-ID 209 of the schema information 122), the device corresponding to “FILE2” is “Sde2” (device name 212 of the DB file information 124), and the logical corresponding to the device of “Sde2” The storage area is “Lun2” of “ST1” (ST-ID 302 and LUN 304 of OS mapping information 114), and the number of HDDs constituting “Lun2” of “ST1” is 10 (the number of HDDs 404 of ST mapping information 172). The processing parallelism of processing block B at this time is 1,536. Processing parallelism per 1HDD processing block B is 1,536 / 10 ≒ 153.

  When processing block A of query 520 is targeted, the object to be accessed is “T1” (access destination Object ID 238 of query plan information 128), and the DB file storing “T1” is “FILE1”. "(File-ID 209 of schema information 122) and the device corresponding to" FILE1 "is" Sdd1 "(device name 212 of DB file information 124), and the logical storage area corresponding to the device of" Sdd1 " Is “Lun1” of “ST1” (ST-ID 302 and LUN 304 of the OS mapping information 114), and the number of HDDs constituting “Lun1” is 10 (the number of HDDs 404 of the ST mapping information 172). The processing parallelism of block A is 100, and 1 HDD of processing block A Processing parallelism of or is 10.

  Next, the analysis / visualization program 195 refers to the system information 196 to acquire the number of simultaneous I / O processes (third input / output process number) per HDD (30 in FIG. 5A), and calculates it in step 736. Compared with the processing parallelism per HDD (step 738), when the number of simultaneous I / O processes per HDD is lower (Yes in step 738), the simultaneous I / O per HDD is compared with the processing parallelism per HDD. The number of O processes is set (step 740). On the other hand, if the number of simultaneous I / O processes per HDD is higher (No in step 738), the processing parallelism per HDD remains unchanged. Subsequently, the analysis / visualization program 195 multiplies the obtained processing parallelism per HDD by the number of HDDs obtained in step 734 to calculate a processing parallelism model prediction value for the entire system (step 742). Hereinafter, the portion described as the process parallelism model predicted value including the drawings means the process parallelism model predicted value of the entire system.

  For example, when processing block B of query 520 is the target, the processing parallelism per HDD calculated in step 736 is 153, the number of simultaneous I / O processes per HDD is 30, and the simultaneous I / O per HDD is 30. Since the number of / O processes is smaller, the processing parallelism per HDD of the processing block B is 30, and 300 obtained by multiplying the number of corresponding HDDs by 10 becomes the processing parallelism model predicted value of the processing block B. If processing block A of query 520 is the target, the processing parallelism per HDD calculated in step 736 is 10, and the number of simultaneous I / O processes per HDD is larger, so that processing block A The processing parallelism per HDD is 10, and 100, which is the result of multiplying the number of corresponding HDDs by 10, is the processing parallelism model predicted value of the processing block A.

  Subsequently, the analysis / visualization program 195 refers to the sequential I / O performance 220 or the random I / O performance 222 of the DB statistical information 126 of the DB monitor information 197, and processes the number of rows of objects accessed by the processing block. The processing time required for execution with the parallelism model predicted value is calculated (step 742). When the processing block B of the query 520 is a target, the scan type of the processing block B is an index scan, the number of selected rows of the object “T2” to be accessed is 100,000, and the processing parallelism model predicted value is 300. Since the random access performance is 10 ms, the processing time of the processing block B is selected row number / processing parallelism model predicted value × access performance = 100,000 / 300 × 10 ms≈3.3 s. When the processing block A of the query 520 is a target, the scan type of the processing block A is an index scan, the number of selected rows of the object “T1” to be accessed is 100, and the processing parallelism model predicted value is 100. From the random access performance of 10 ms, the processing time of the processing block A is selected row number / processing parallelism model predicted value × access performance = 100/100 × 10 ms≈10 ms.

  With the above processing, the processing behavior predicted when the query 520 is executed is that processing of the processing block B is executed for 3.3 s with a processing parallelism of 30 and then processing of the processing block A is executed with a processing parallelism of 10 for 10 ms. It will be to.

  FIG. 9 is a flowchart of the process parallelism actual value calculation process according to the first embodiment. In the first embodiment, the number of tags for each HDD is calculated as the measured processing parallelism value. Specifically, the analysis / visualization program 195 reads event-based ST processing information one by one, and increments or decrements the number of tags of the corresponding HDD according to the event (in the case of a disk I / O start event). Increment and decrement in case of disk I / O end event). Then, the analysis / visualization program 195 acquires the number of tags for each processing block of all HDDs at that time as a measured processing parallelism value at every predetermined sampling interval.

  The analysis / visualization program 195 first clears and initializes the number of tags of all HDDs to 0 (step 801), reads 1 entry of the ST processing information 174 in the ST monitor information 199 (step 802), and reads all entries. When it is determined whether or not all entries have been read (step 804: No), the processing from step 808 is executed while all entries have been read. In (Yes in step 804), the process parallelism actual value calculation process ends (step 806).

  The analysis / visualization program 195 determines whether the time stamp of one entry of the ST processing information 174 read in step 802 exceeds a predetermined sampling interval for actual measurement calculation (step 808), and the sampling interval for actual measurement calculation. If it exceeds (Yes in Step 808), the number of tags for each processing block of all HDDs (variable used for processing) at that time is acquired as the number of tags at that time (Step 810). On the other hand, if the sampling interval for actual value calculation is not exceeded (No in step 808), nothing is done. Here, the sampling interval for calculating the actual measurement value may be given as a parameter when the analysis / visualization program 195 is started, or may be set by the system administrator before starting the analysis.

  Subsequently, when the event of one entry of the ST processing information 174 read in step 802 is a “Disk I / O start” event, the analysis / visualization program 195 executes the OS processing of the OS monitor information 198. The number of tags for each processing block of the disk 178 to be accessed is specified by referring to the DB processing information 130 of the information 116 and the DB monitor information 197, specifying the query that issued the I / O of the event, and the processing block in the query Is incremented (step 812). For example, in the case of the “disk I / O start” event of entry number 3 in the ST processing information 174, the I / O identifier (IO-ID) is “IO1”. The OS processing information 116 entry and the DB processing information 130 entry having the same identifier “IO1” are searched and found, respectively, and the query “Q1” that issued the I / O and the processing block “A” in the query are specified. .

  Subsequently, when the event of one entry of the ST processing information 174 read in step 802 is a “Disk I / O end” event, the analysis / visualization program 195 executes the OS processing of the OS monitor information 198. Referring to the information 116 and the DB processing information 130 of the DB monitor information 197, the query that issued the I / O of the event and the processing block in the query are specified (the specification procedure is the same as in step 812), and the access destination The number of tags for each processing block of the disk 178 is decremented (step 814). Then, the analysis / visualization program 195 returns to step 802 and reads the next entry of the ST processing information 174.

  According to the above processing parallelism actual value calculation processing, the number of tags for each processing block for each disk 178 can be calculated at the actual value sampling interval, and the actual processing behavior of the DBMS 120 when the query is actually executed is specified. can do.

  Subsequently, the graph displayed by the analysis / visualization program 195 based on the calculated process parallelism model predicted value and the process parallelism actual measurement value will be described.

  FIG. 10A is a diagram illustrating an example of a process parallelism model predicted value graph according to the first embodiment. A graph 900 illustrated in FIG. 10A is a graph in which the vertical axis represents the process parallelism model predicted value and the horizontal axis represents time, and the process parallelism model predicted value calculated in the process parallelism model predicted value calculation process is expressed in time series. It is drawn. In the graph 900, the graph line 902 indicates the processing parallelism model predicted value, the time 0 to t1 is the processing parallelism model predicted value of the processing block B, and the time t1 to t2 is the processing parallelism model of the processing block A. It is a predicted value.

  FIG. 10B is a diagram illustrating a first example of a graph of the process parallelism model predicted value and the process parallelism actual measurement value according to the first embodiment. A graph 910 illustrated in FIG. 10B is a graph in which the vertical axis represents processing parallelism (process parallelism model predicted value and processing parallelism actual measurement value) and the horizontal axis represents time, and is calculated by the processing parallelism model predicted value calculation process. The process parallelism model predicted value and the process parallelism actual value calculated by the process parallelism actual value calculation process are drawn in time series. In the graph 910, the process parallelism model predicted value is indicated by a graph line 902, and the process parallelism actual measurement value is colored and overlaid. According to the example shown in the graph 910, it can be seen that the processing parallelism model predicted value and the processing parallelism actual measurement value are suitable for both the processing block A and the processing block B in the range of time t0 to t2. , It can be determined that it is operating according to the processing behavior predicted based on the model.

  FIG. 10C is a diagram illustrating a second example of the graph of the process parallelism model predicted value and the process parallelism actual measurement value according to the first embodiment. A graph 920 illustrated in FIG. 10C is a graph similar to the graph 910. According to the example shown in the graph 920, the process parallelism model predicted value and the process parallelism actual measurement value are incompatible within a part of the time t0 to t3 of the process block B, and the process block B has the time 0 to 0. It can be seen that the range of t0 is compatible and the range of the processing block A is compatible. According to this graph 920, the DBMS 120 can determine that it does not operate according to the processing behavior predicted based on the model. In this case, it is considered that a problem has occurred, and the occurrence time of the problem is likely to be t0. This can be easily determined.

  As shown in the graphs 910 and 920, the processing parallelism model predicted value and the processing parallelism actual measurement value are displayed by overlapping the processing parallelism model predicted value and the processing parallelism actual measurement graph for each processing block. Can be easily compared. In the graphs of the measured parallelism values in the graphs 910 and 920, I / Os generated from systems and programs other than the DBMS 120 are displayed in different display forms (for example, the graph color is changed to gray). It is also good. In this case, the I / O from the system and program other than the DBMS 120 is the “IO-ID” included in the DB processing information 130, the OS processing information 116, and the ST processing information 174 in the processing parallelism actual value calculation processing 800 described above. ”Can be specified as an I / O that is not linked. Thereby, it is possible to distinguish between I / O from the DBMS 120 and I / O from other systems and programs, and it is possible to analyze the processing behavior of only the DBMS 120.

  Next, an example of a screen displayed by the analysis / visualization program 195 will be described.

  FIG. 11A is a diagram illustrating a screen display example including a graph of the processing parallelism model predicted value according to the first embodiment. When the analysis / visualization program 195 is executed, a window 1000 is displayed on the display 144 as shown in FIG. 11A. In the window 1000, a menu bar 1001, a display target HDD selection box 1002, a model predicted value display selection button 1004, an actual value display selection button 1006, a graph drawing area 1008, and a window close button 1012 are arranged. Thus, it is possible to perform an instruction operation on each button.

  In the menu bar 1001, a menu for giving instructions for starting analysis, switching screens, and ending a program is prepared. In a display target HDD selection box 1002, an HDD that is a target of a graph to be drawn in the graph drawing area 1008 can be selected. When “ALL” is selected in the display target HDD selection box 1002, the analysis / visualization program 195 draws a graph of the processing parallelism of the entire system in the graph drawing area 1008. When the model predicted value display button 1004 is selected and instructed, the analysis / visualization program 195 sets the model predicted value display button 1004 to the selected state (in the figure, the bold frame display indicates the selected state), and the processing parallelism model prediction A graph of the process parallelism model predicted value calculated by the value calculation process is drawn in the graph drawing area 1008. When the actual value display button 1006 is selected and instructed, the analysis / visualization program 195 places the actual value display button 1006 in a selected state (in the figure, a bold frame indicates the selected state), and the processing parallelism actual value A graph of the measured parallelism value calculated in the calculation process is drawn in the graph drawing area 1008. In the example of FIG. 11A, only the process parallelism model predicted value is displayed.

  FIG. 11B is a diagram illustrating a screen display example including a graph of the process parallelism model predicted value and the process parallelism actual measurement value according to the first embodiment. In the example of FIG. 11B, the model predicted value display button 1004 and the measured value display button 1006 are selected, and the processing parallelism model predicted value and the processed parallelism measured value are superimposed in the graph drawing area 1008. A graph is displayed. When the mouse cursor is placed on the graph drawn in the graph drawing area 1008, the analysis / visualization program 195 displays information on the location (time stamp, query identifier, processing block identifier, processing parallel number, etc.). ) Is displayed with a tool tip 1010. When the close button 1012 is clicked with the mouse, the analysis / visualization program 195 closes the window 1000 and ends the processing.

  By operating the screen displayed by the analysis / visualization program 195, the system administrator can display the graph of the processing parallelism model predicted value and the processing parallelism actual measurement value, and predicting based on the model of the DBMS 120 With respect to the processing behavior, it is possible to easily and appropriately determine how it actually operates and whether a problem has occurred.

  Next, a display screen used for investigating the cause of the problem will be described.

  FIG. 11C is a diagram illustrating a screen display example including a graph regarding the staying number according to the first embodiment.

  FIG. 11C is a screen including a graph displaying the staying status for each place (element: process staying place) where processing staying in the computer system can be considered. Examples of the processing staying location include the inside of the DBMS 120 of the computer 100 (processing element by the DBMS 120), the inside of the OS 112 (processing element by the OS 112), the HBA that is an example of the I / F (1) 136, the communication network 140, and the storage device 150. There are components of the controller 154 and the HDD 178.

  Here, first, a process of acquiring the staying status of each staying place by the analysis / visualization program 195 will be described. The analysis / visualization program 195 refers to the DB processing information 130 of the DB monitor information 197, the OS processing information 116 of the OS monitor information 198, and the ST processing information 174 of the ST monitor information 199, and an event corresponding to each staying location (for example, The retention in the DBMS 120 is incremented or decremented based on the “IO-Submit” event to the “IO-GetEvent” event in the DB processing information 130), and the processing parallelism actual value is calculated. Similar to the processing, the number of stays at each measured sampling time can be obtained by determining and obtaining the number of stays at each stay location at the measured sampling time. The analysis / visualization program 195 displays the number of stays as a time-series graph using the number of stays for each measured sampling time for each stay location.

  The screen displayed in FIG. 11C may be switched and displayed when the screen switching menu in the menu bar 1001 of the window 1000 is selected, or may be automatically identified and displayed when a problem occurs. May be. In the screen shown in FIG. 11C, the window 1000 includes a menu bar 1001, a close button 1012, and a graph drawing area 1010. In the graph drawing area 1010, the analysis / visualization program 195 draws a time-series graph of the number of stays at each stay location with the vertical axis as the stay number and the horizontal axis as the time. In FIG. 11C, the staying status at the staying location 1 is represented by a graph 1012, the staying status at the staying location 2 is represented by a graph 1014, and the staying status at the staying location 3 is represented by a graph 1016. For example, when it is found from the graph of the process parallelism model predicted value and the process parallelism actual measurement value that a problem has occurred at time t0, the situation of each staying location at time t0 is grasped from the screen of FIG. 11C. The cause can be easily identified. For example, in the screen example of FIG. 11C, it can be seen that the staying state of the staying portion 2 has changed greatly at time t0, so that the cause of the problem can be easily identified in the staying portion 2.

  Next, another screen example displayed by the analysis / visualization program 195 will be described.

  FIG. 12A is a diagram illustrating a screen display example including a graph indicating conformity / nonconformity between the model predicted value and the actual measurement value of the processing parallelism in the HDD according to the first embodiment. FIG. 12B is a diagram illustrating a screen display example indicating conformity / nonconformity between the model predicted value of the processing parallelism and the actual measurement value for each HDD according to the first embodiment.

  In the screens shown in FIGS. 11A and 11B, the theoretical value of processing parallelism and the actual processing parallelism value for each HDD can be displayed. If the number of HDDs is small, the system administrator can display a graph of the process parallelism model predicted value and the process parallelism actual measurement value for each HDD, and determine whether each HDD has operated properly. it can. However, in the case of a computer system having hundreds to thousands of HDDs, it takes a lot of time to do the same thing for each HDD.

  Therefore, the analysis / visualization program 195 automatically calculates the nonconformity rate of the processing parallelism (HDD tag number) for each HDD and displays the result, thereby reducing the analysis man-hours for the system administrator, and the behavior / performance Facilitates analysis. Here, the nonconformity rate of the processing parallelism is a ratio of the nonconformity time to the entire processing time of the query. For example, if the relationship between the processing parallelism model predicted value and the processing parallelism actual measurement value of a certain HDD is as shown in the graph of FIG. 10C, the entire processing time is t4, and the nonconforming time is t3−. t0, and the nonconformity rate can be expressed as t4 / (t3-t0). Here, the analysis / visualization program 195 compares the process parallelism model predicted value and the process parallelism actual measurement value, for example, at each actual measurement value sampling interval, and determines that the two match if they match. If not, it is determined as nonconforming. If the system administrator sets a threshold value for determination of conformity / non-conformity for the analysis / visualization program 195 and the measured parallelism value for the predicted processing parallelism model value falls within the threshold value range. You may make it determine with a conformity.

  FIG. 12A is an example of a screen that displays the number of conforming / nonconforming HDDs between the predicted processing parallelism model value and the actual processing parallelism value in all HDDs of the computer system over time. The window 1000 includes a menu bar 1001, a close button 1012, and a graph drawing area 1010. In the graph drawing area 1010, a graph of the number of conforming / nonconforming HDDs is drawn with the vertical axis representing the number of HDDs and the horizontal axis representing the elapsed time. In the screen shown in FIG. 12A, a line 1102 indicates all HDDs, a graph 1104 indicates the number of compatible HDDs, and a graph 1106 indicates the number of non-compatible HDDs. The determination as to whether each HDD is compatible or not compatible in the entire processing may be performed by determining that an HDD with a non-conformance rate of 0% is compatible, or the parameters given by the system administrator to the analysis / visualization program 195 ( For example, it may be determined that the nonconformity rate is 20% or less. With the screen shown in FIG. 12A, it is possible to easily identify how much HDD is operating with the processing behavior predicted based on the model in the entire processing by the DBMS 120.

  FIG. 12B is a screen in which the conformity / nonconformity state between the processing parallelism model predicted value and the processing parallelism actual measurement value in all HDDs of the computer system is displayed in a matrix in the graph drawing area 1010.

  In the figure, each rectangular portion in the matrix corresponds to one HDD. This rectangular portion indicates an HDD in which the processing parallel power model predicted value and the processing parallel degree actual measurement value are compatible, an HDD in which the processing parallel power model predicted value and the processing parallel degree actual measurement value are incompatible, and an unaccessed in the query Are displayed in different display modes (for example, different colors). Here, a rectangle 1110 indicates a compatible HDD, a rectangle 1112 indicates a non-compatible HDD, and a rectangle 1114 indicates an unaccessed HDD. As for the method of determining whether each HDD is conforming or nonconforming by the analysis / visualization program 195, it is good to determine that the HDD having a nonconformity rate of 0% is conforming as described above, and the system administrator analyzes and visualizes it. A parameter or less given to the program 195 (for example, nonconformity rate 20%) or less may be determined as conforming.

  In FIG. 12B, all HDDs are shown in one matrix, but these groups may be displayed in one matrix, such as for each storage device 150 or for each RAID group. When one rectangular portion in the matrix is selected by an operation such as a mouse click, the analysis / visualization program 195 performs processing parallelism model as shown in FIG. 11B for the HDD corresponding to the selected rectangular portion. You may make it display the graph of a predicted value and a process parallelism measured value. As a result, it is possible to easily and in detail analyze the conformity / nonconformity state between the process parallelism model predicted value and the process parallelism actual measurement value.

  FIG. 12C is a diagram illustrating a screen display example in which HDDs having a high mismatch rate between the model predicted value and the actual measurement value of the processing parallelism according to the first embodiment are sequentially illustrated.

  In the window 1000 shown in FIG. 12C, the analysis / visualization program 195 ranks and displays HDDs having a high mismatch rate between the processing parallelism model predicted value and the processing parallelism actual measurement value in the graph display area 1010 in order. In the graph display area 1010, a ranking number is displayed in the area 1120, an identifier (ST-ID) for identifying the storage device 150 and an identifier (HDD-ID) for identifying the HDD 178 are displayed in the area 1122, and an area 1124 is displayed. Shows the nonconformity rate.

  In FIG. 12C, it is displayed that the HDD “HDD45” of the storage device “ST1” is first in the nonconformity rate 32.5%, and the HDD “HDD6” of the storage device “ST1” is the nonconformance rate 31.8%. The second place is displayed, and the HDD “HDD74” of the storage device “ST1” is displayed in the third place with a nonconformity ratio of 31.4%.

  In the graph display area 1010, a detailed display button 1126 is displayed at a position corresponding to each entry in the ranking. When an operation such as a mouse click is performed on the detail display button 1126, a graph of the processing parallelism model predicted value and the parallel processing degree actual measurement value for the corresponding HDD is displayed. Thereby, it is possible to analyze in detail the conformity / nonconformity state between the processing parallelism model predicted value and the processing parallelism actual measurement value for each HDD displayed as having a high nonconformity rate.

  Further, another screen example displayed by the analysis / visualization program 195 will be described.

  According to the screens shown in FIGS. 11A and 11B, the processing parallelism model predicted value and the processing parallelism actual measurement value in the computer system are displayed, and whether the DBMS 120 is operating according to the processing behavior predicted based on the model is a problem. It can be easily determined whether or not it has occurred, and the location where the problem has occurred can be easily identified by the screen shown in FIG. 11C. However, it is impossible to investigate and analyze the actual behavior of the DBMS 120 in more detail only with these screens.

  Accordingly, the analysis / visualization program 195 analyzes the system information 196, the DB monitor information 197, the OS monitor information 198, and the ST monitor information 199, and displays the processing status of each thread in the DBMS 120 on the screen, thereby the actual behavior of the DBMS 120. To facilitate the investigation and analysis of

  FIG. 13A is a diagram illustrating a screen display example including a table accessed by the DBMS according to the first embodiment and a graph of a statistical value of the number of tasks for each index. A graph 1200 shown in FIG. 13A is an object in time series indicating the number of tasks for each accessed object in the graph drawing area 1010 of the window 1000, with the number of tasks for each object accessing the vertical axis and the horizontal axis for time. Each figure is drawn with a different display mode (for example, the color of the graph). With this graph 1200, it is possible to easily specify which task is accessing which object in each time zone.

  FIG. 13B is a diagram illustrating a first screen display example including a graph of statistical values for each task state according to the first embodiment. A graph 1210 shown in FIG. 13B is displayed in a graph drawing area 1010 of the window 1000, in which the number of tasks for each task state is displayed in time series for each task state, with the vertical axis representing the number of tasks for each task state and the horizontal axis representing time. For example, the graph is drawn with a different color). As the task state, for example, a DB process is being executed, an I / O issuance state, a CPU allocation waiting state, a DB buffer I / O completion waiting state, a lock holding state, and a lock acquisition waiting state are taken. In the example shown by the graph 1210, the graph 1212 shows the number of tasks waiting for I / O completion of the DB buffer, the graph 1214 shows the number of tasks waiting for lock acquisition, and the graph 1216 shows the state waiting for CPU allocation. The graph 1218 shows the number of tasks in an I / O issuance state. According to the graph 1210, it is possible to easily identify whether the task is operating at the upper limit that can be executed in each time slot, and what task state it is in.

  FIG. 13C is a diagram illustrating a second screen display example including a graph indicating the state of each task according to the first embodiment. A graph 1220 shown in FIG. 13C is displayed in the graph drawing area 1010 of the window 1000 in a display form for each task state, where the vertical axis indicates a task (one horizontal line indicates the state of one task) and the horizontal axis indicates time. For example, the graph is drawn with a different color). Of the task states, only one of the states other than the lock holding state is taken, but the lock holding state overlaps with other states (for example, issuing an I / O while holding the lock) Therefore, the graph in the lock holding state is drawn below the other graphs. According to the graph 1220, it is possible to easily specify the state of each task in each time slot.

  In the first embodiment, the processing parallelism model predicted value is the number of selected rows, the number of kernel threads, the number of tasks, the number of HBA simultaneous I / O processes, the simultaneous I / O processing of the storage controller corresponding to the key value of the index to be referenced. The number is calculated based on the number of simultaneous I / O processes per HDD. However, although not described in the first embodiment, the queue length on the I / O path in the VM (Virtual Machine) in which the OS is multistaged, the file system in the OS, the scheduler, the device driver, etc., or on the network In addition to the throttling by the QOS control, the rate limiting factor of the processing parallelism can be considered. In the first embodiment, means for acquiring information related to the part that is a rate-limiting factor may be provided, and calculation may be performed including them when calculating the processing parallelism.

  In the first embodiment, the case where the DBMS 120 executes one query has been described, but when the DBMS 120 executes a plurality of queries at the same time, the processing parallelism model predicted value and the processing parallelism actual measurement value are calculated in the same manner. By displaying the result on the screen, it is possible to analyze the processing behavior when multiple queries are executed. In this case, a plurality of queries can be identified by “QID” included in the query plan information 128 and the DB processing information 130. Regarding the calculation of the process parallelism model predicted value, the process parallelism model predicted value calculation process 610 is executed for each query, and the process model predicted value of the entire system can be calculated by adding them. Regarding the calculation of the processing parallelism actual measurement value, it can be calculated by counting each query with the above-mentioned “QID”. In addition, when displaying the processing parallelism measured value when executing multiple queries on the screen, it is possible to show how each query was executed by changing the display form (for example, graph color) for each query. It becomes possible to analyze.

  FIG. 14 is a diagram illustrating an example of a graph of the process parallelism model predicted value and the process parallelism actual measurement value when a plurality of queries (query X and query Y in this example) are executed. In the graph 1400 illustrated in FIG. 14, a value obtained by adding the processing parallelism model predicted values of the query X and the query Y is displayed as a graph line 902 of the processing parallelism model predicted value, and the processing parallelism actual measurement value 1402 of the query X is displayed. And the query parallel processing actual measurement value 1404 of the query Y are displayed in different colors.

  In the first embodiment, for example, an I / O path abstraction layer such as LVM (Logical Volume Manager) has not been described. However, a processing parallelism model predicted value and processing in consideration of the I / O path abstraction layer are described. It is also possible to analyze the processing behavior of the DBMS 120 by calculating the measured parallelism value and displaying the result. FIG. 15A and FIG. 15B are configuration diagrams of an example of logical volume information and volume group information when LVM is used. The logical volume information 1500 is information related to mapping between a logical volume that is a logical storage area and a volume group in which the logical volume is created, and has an entry for each logical volume. Each entry has a field 1502 for registering a logical volume name for identifying a logical volume, and a field 1504 for registering a volume group name for identifying a volume group in which the logical volume is created. The volume group information 1510 is information relating to mapping between a volume group and physical volumes (devices) constituting the volume group, and has an entry for each volume group. Each entry has a field 1512 for registering a volume group name for identifying a volume group, and a field 1514 for registering a physical volume (device) constituting the volume group. In this case, the DB file is created on the logical volume, and the created logical volume name is registered in the field 212 for registering the device name of the DB file information 124 in FIG. 2B. In the process of calculating the process parallelism model predicted value and the process parallelism actual measurement value, the logical volume information 1500 and the volume group 1510 are referenced to specify the HDD to be accessed when the DBMS 120 executes the DB process.

  In the first embodiment, the behavior of the DBMS 120 is focused on the processing parallelism model predicted value value (for example, the model predicted value of the number of HDD tags) and the processing parallelism measured value (for example, the measured value of the HDD tag number). Explains how to analyze performance. For example, replace performance with power consumption, and estimate the power consumption model when a query is executed on a computer system, and measure the power consumption when the query is actually executed It is also possible to calculate the value and use it as power consumption monitoring. In particular, in a large-scale system environment, an enormous amount of power is consumed. Therefore, it is useful to calculate a power consumption model prediction value in advance and present it to the system administrator. Further, by displaying the power consumption model predicted value and the power consumption actual measurement value in a comparable manner, the power consumption can be easily analyzed.

  As described above, according to the first embodiment, the system administrator can easily determine whether the DBMS 120 is operating according to the processing behavior predicted based on the model or not. When it is not operating, it is considered that a problem has occurred, and the cause of the problem can be easily identified. Furthermore, the actual behavior of the DBMS 120 can be easily investigated and analyzed, and the time required for the behavior / performance analysis of the DBMS 120 can be greatly reduced.

  Example 2 will be described below. At that time, differences from the first embodiment will be mainly described, and description of common points with the first embodiment will be omitted or simplified.

  In the first embodiment, the analysis / visualization program 195 calculates the model predicted value and displays the model predicted value and the actual measurement value on the display 144. In the second embodiment, the analysis / visualization program 195 displays the model predicted value. It is not always necessary to calculate and display. In the second embodiment, as the query execution progresses, the DB monitor information output unit 132 outputs event-driven information related to task execution as DB processing information 2200 and 2400, which will be described later, and the analysis / visualization program 195 displays the information. Is displayed and a graph showing the task execution status is displayed on the screen.

Of course, Example 1 and Example 2 can also be combined. In the combination of Example 1 and Example 2, for example, the following flow can be considered.
(A) The analysis / visualization program 195 calculates a model predicted value and an actual measurement value of the processing behavior of the DBMS 120 based on the acquired various monitor information, and displays these graphs (FIGS. 11A to 11C, as necessary ( For example, in response to a request from the system administrator, FIGS. 12A-12C) are displayed on the display 144.
(B) The system administrator looks at these graphs and determines whether a problem has occurred in the processing of the DBMS 120.

  When it is determined in (B) that a problem has occurred, the system administrator can identify the problem that has occurred by performing the following operation. As a problem to be identified, for example, the DBMS 1202 can execute tasks in parallel even though the tasks can be executed in parallel.

  For example, the system administrator causes the analysis / visualization program 195 to display graphs relating to the behavior of the DBMS 120 (for example, FIGS. 13A to 13C) by operating a menu on the screen displayed on the display 144. As a result, the system administrator can analyze the thread execution state, the object access state, and the like.

  Further, for example, the system administrator operates the menu of the screen displayed on the display 144 to cause the analysis / visualization program 195 to display the visualization window 2500 (FIG. 25), and the analysis / visualization program 195 Then, a task behavior visualization tree (for example, FIG. 26 or FIG. 30) and / or a data structure visualization tree (for example, FIG. 27) is rendered on the visualization window 2500. The display mode of each node in the task behavior visualization tree and the data structure visualization tree changes according to the progress of execution of the query (according to the event type in the DB processing information shown in FIGS. 22 and 24). As a result, the system administrator can identify a problem that has occurred during execution of the query.

  Hereinafter, Example 2 will be described in detail.

  In the second embodiment, for example, there are an index IA, a table A, an index IB, and a table B as shown in FIG. The index IA is stretched on the column A2 of Table A, and the index IB is stretched on the column B1 of Table B.

  When the DBMS 120 receives the query shown in FIG. 17A for the schema group shown in FIG. 16, the DBMS 120 generates the query plan shown in FIG. 17B and acquires the query result shown in FIG. 17C as the execution result.

  In the schema group shown in FIG. 16, the data structure of index IA and table A is as shown in FIG. 18, and the data structure of index IB and table B is as shown in FIG. It is assumed that the structure is 18 and 19, “P” represents a DB page. Hereinafter, “Pxxx” (xxx is a three-digit integer) is expressed as DB page #xxx. For example, according to FIG. 18, the value in the record including the value of “ABB” or more as the value of the column A2 refers to the DB page # 103 according to the page reference from the DB page # 101, and the DB page # 103. Can be acquired from the DB pages # 105 and 106 by referring to the DB pages # 105 and 106 according to the record reference.

  The index has a tree structure here, but may have another structure such as a hash.

  The record reference is a physical reference (for example, a page ID and a storage position in the page) here, but may be a logical reference (a unique primary key). In that case, an index entry can be obtained from the secondary index and the main index can be searched to obtain a record.

  The data structure of the table is a flat file, but it may be a tree structure or the like like the index as in the index-organized table.

  The information representing the above data structure is usually so large that it is difficult to store the information in the memory 110 of the computer 100. Therefore, the DBMS 120 creates a data structure summary table that is a summary of the information, and stores the data structure summary table in the memory 110.

  FIG. 20 shows an example of the data structure summary table 2000. This table 2000 includes, for each DB page, a page ID, an object ID (an ID of a schema having a value stored in the DB page), and a parent DB page ID (a parent DB page of a DB page (reference source DB page). ) ID) and key range (upper and lower limits of usable key values).

  Assume that there are 25 tasks # 1 to # 25.

  In the case of a conventional DBMS, 25 tasks are sequentially executed as shown in FIG. DB processing information representing a set of records in which the execution history is described is, for example, DB processing information 2200 shown in FIG. The DB processing information 2200 has one or more records. One record corresponds to one task execution history, and includes an entry number (history number), a time stamp (task execution time or number), and a thread ID (ID of a thread having a executed task). And an event type (type of an event performed by task execution), a task ID (ID of the executed task), and a parent task ID (ID of the parent task of the executed task). The DB processing information 2200 is a modification of the DB processing information 130 shown in FIG. 2E.

  However, the DBMS 120 described above executes tasks in parallel. For example, the tasks # 1 to # 25 are executed as shown in FIG. More specifically, the DBMS 120 generates and executes task # 1, and when executing the next DB operation based on the execution result of the DB operation corresponding to the executed task # 1, Based on the newly generated tasks # 2 and # 16, the newly generated tasks # 2 and # 16 are executed in parallel. As a result, as can be seen by comparing FIG. 21 and FIG. 23, the time required for executing the query can be shortened. Note that the DB processing information when the tasks # 1 to # 25 are executed as illustrated in FIG. 23 is, for example, as the DB processing information 2400 illustrated in FIG. The DB processing information 2400 is stored in the memory 110 of the computer 100.

  As can be seen from the comparison between the records of entry numbers 6 and 7 in FIG. 22 and the records of entry numbers 6 and 7 in FIG. 24, the conventional DBMS sequentially executes operations corresponding to task # 2. However, the DBMS 120 according to the second embodiment executes the operation corresponding to the task # 2 and the operation corresponding to the task # 16 in parallel.

  Even if some problem occurs in the behavior of the conventional DBMS, it is considered that it is not difficult to identify the problem because the tasks are sequentially performed in the order of occurrence.

  However, if any problem occurs in the behavior of the DBMS that executes tasks in parallel as in the second embodiment (and the first embodiment), it is not determined in what order the tasks are executed. It is difficult to identify the problem.

  As a method for solving this problem, it is conceivable to visualize the behavior of the DBMS.

  However, even if the behavior of the DBMS (how the DBMS operates on the computer) is simply visualized, the system administrator cannot always identify the problem. For example, the analysis / visualization program can visualize the behavior of the DBMS by displaying the DB processing information 2400 shown in FIG. 24 as text. However, in order to identify the problem, the system administrator needs to follow the records in the displayed DB processing information in chronological order, and the burden for identifying the problem is large.

Thus, in the second embodiment, the analysis / visualization program 195 acquires the DB processing information 2400 from the computer 100 and stores the acquired DB processing information 2400 in the memory 192. The analysis / visualization program 195 determines the behavior of the DBMS 120 based on the DB processing information 2400 in the memory 192 as follows:
(A) drawing a task object which is an object representing the executed task;
(B) The display mode of the task object is changed in accordance with the change in the state of the task represented by the task object (the type of event performed by the task execution).
Visualize according to For example, the task object may be a node in the tree, or may be a fragment area (for example, a sector area) obtained by equally dividing a predetermined area (for example, a circle) by the number of executed tasks. good. Also, various display modes such as color, pattern, brightness, task object (for example, node) size, line thickness, or change from non-display to display are adopted as display modes that change depending on the task status. can do. The analysis / visualization program 195 may change the display mode of the task object by animation. Further, the analysis / visualization program 195 may draw an object representing a parent-child relationship of tasks (for example, an edge (link) connecting task objects) based on the DB processing information 2400.

  As the DB size increases, the number of task objects increases, and if the analysis / visualization program 195 performs the above drawing, the drawing contents become overcrowded and the user (analyst) interprets the behavior. There is a possibility that you do not know what to do. Therefore, the analysis / visualization program 195 may be able to switch between a display form for displaying all of the task objects and a display form for displaying a part of the designated task objects according to a user instruction. In addition, when selecting a part of the task object to be displayed, the range of the task object to be displayed may be selected by, for example, setting the mouse pointer to the task object to be selected and zooming in and out from that point. .

  Several specific examples will be described below.

  The analysis / visualization program 195 displays a visualization window 2500 as shown in FIG. 25 on the display 144 when information representing an operation for displaying the visualization window is received from the input / output device 188.

  The visualization window 2500 includes a task display area 2501 in which a task behavior visualization tree is displayed, and an object display area in which an object data structure visualization tree is displayed. As the object display area, for example, one or more index display areas 2502A and 2502B in which one or more data structure visualization trees respectively corresponding to one or more indexes are displayed, and one or more tables corresponding to one or more tables, respectively. There are one or more table display areas 2503A and 2503B in which a data structure visualization tree is displayed. The index display area and the table display area referenced from the index are arranged along a predetermined direction (for example, the vertical direction).

  The visualization window 2500 includes a time stamp slider 2504, a zoom-in button 2505, and a zoom-out button 2506. The time stamp slider 2504 is an example of a time stamp adjustment tool. One end of a bar 2505 on which the slider 2504 runs is the execution start time of the root task (first task) (the time indicated by the time stamp in the first record of the DB processing information 2400). The other end of the bar 2505 is the time when the DB operation corresponding to the end task (the last task) is completed (the time indicated by the time stamp in the end record of the DB processing information 2400).

  FIG. 26 is an example of a task behavior visualization tree drawn in the task display area 2501.

  The task behavior visualization tree is a graph in which the task is a node and the parent-child relationship of the task is an edge (link).

  For example, 25 nodes corresponding to 25 tasks # 1 to # 25 are arranged along a predetermined direction (for example, the vertical direction) from the oldest time stamp to the newest time stamp. In FIG. 26, for example, the root task (the task with the oldest time stamp) is at the top, and the end task (the task with the oldest time stamp) is at the bottom. The analysis / visualization program 195 equally divides the vertical coordinate range in the display range of the task behavior visualization tree by the number of nodes arranged in the vertical direction, and draws the node at each equally divided coordinate position. In this division, the analysis / visualization program 195 considers two or more nodes having substantially the same time stamp of the event type “task start” as one node. Therefore, it can be seen from this tree that two or more tasks respectively corresponding to two or more nodes having the same vertical coordinate are started at substantially the same timing.

  In the task behavior visualization tree, nodes equidistant from the node corresponding to the root task are arranged at equal intervals along the horizontal direction.

  State types corresponding to the display mode of nodes and edges in the task behavior visualization tree include, for example, an unstarted task, an IO task (task whose event type is “Task Start” or “IO-Submit”), and an OP task There are four types of tasks (tasks whose event type is “IO-GetEvent” or “OP Start”) and OP completion tasks (tasks whose event type is “OP End”). In the initial state, the analysis / visualization program 195 draws all nodes and edges in a display mode corresponding to the unstarted task, and then displays the node and edge display modes corresponding to the task state at the specified time. Change to embodiment. The tree shown in FIG. 26 represents the task execution state at the time stamp “103.00”. According to the DB processing information 2400 of FIG. 24, the event type of task # 1 is “OP End” with the time stamp “103.00”, and the event type of tasks # 2 and 16 is “IO-Submit”. And other tasks have not started yet. For this reason, the analysis / visualization program 195 changes the display mode of the node and edge corresponding to the task # 1 to the display mode corresponding to the OP completion task (for example, an intermediate display mode (IO task, Change the display mode of the nodes and edges corresponding to the tasks # 2 and 16 to the display mode corresponding to the task during IO, and the nodes and The display mode of the edge remains the display mode corresponding to the unstarted task.

  FIG. 27 is an example of a data structure visualization tree drawn in the schema display areas 2502A, 2502B, 2503A, and 2503B.

  The data structure visualization tree is a graph in which a DB page is a node and a dependency between pages on the data structure is an edge (link).

  The analysis / visualization program 195 acquires the data structure summary table 2000 from the computer 100 and stores the acquired data structure summary table 2000 in the memory 192. The analysis / visualization program 195 draws a data structure visualization tree based on the DB processing information 2400 and the data structure summary table 2000 in the memory 192.

  The parent-child relationship between DB pages on the data structure may be a link or the like in the case of a link array. When the data structure is a flat file, the analysis / visualization program 195 may draw the nodes corresponding to the DB page by arranging them in the page address order without drawing them as a graph structure. The analysis / visualization program 195 may draw the entire database instead of the data structure by arranging the nodes in the order of the DB page addresses.

  Nodes equidistant from the node corresponding to the root page of the data structure are arranged at equal intervals in the horizontal direction.

  As status types corresponding to display modes of nodes and edges in the data structure visualization tree, for example, an unIO page (a DB page whose corresponding task is an unstarted task), an IO-completed page (a corresponding task is an in-IO task) DB page), OP page (DB page whose corresponding task is an OP task), and OP completed page (DB page whose corresponding task is an OP completion task). In the initial state, the analysis / visualization program 195 draws all nodes and edges in a display mode corresponding to an unIO page, and then changes the display mode of the nodes and edges depending on the task state at the specified time. Change to the display mode corresponding to the page state. The tree shown in FIG. 27 represents a state at the time stamp “103.00”. According to the DB processing information 2400 shown in FIG. 24, the DB page # 101 is a DB page referred to by the task # 1 that is the OP completion task with the time stamp “103.00”, and the DB pages # 102 and 103 Are DB pages that are referred to by tasks # 2 and 16 that are in-IO tasks, and other DB pages are DB pages that are referred to by tasks that are unstarted tasks. For this reason, the analysis / visualization program 195 changes the display mode of the node and edge corresponding to the DB page # 101 to the display mode corresponding to the OP-completed page (for example, an intermediate display mode (IO-completed page in animation)). Change the display mode of the nodes and edges corresponding to the DB pages # 102 and 103 to the display mode corresponding to the IO completed page, and support other DB pages. The display mode of the nodes and edges to be kept is the display mode of the non-IO page.

  As the specified time changes (as the slider 2504 moves), the analysis / visualization program 195 changes the display mode of the nodes and edges of the various trees to the event type (or earlier than the specified time before the specified time after the transition). Change the display mode according to the event type. For example, when the designated time is changed from the first time “103.00” to the second time (for example, a certain time in the future after the first time), the analysis / visualization program 195 stores the DB processing information 2400. Based on this, the display of the task execution visualization tree is changed from the display shown in FIG. 26 to the display shown in FIG. 28 (for example, changed by animation), and the display of each data structure visualization tree is shown in FIG. The display is changed to the display shown in FIG. 29 (for example, it is changed by animation). Specifically, the analysis / visualization program 195 switches the record reference destination in the DB processing information 2400 as the specified time changes, and the value (especially the event type) in the record referred to when the record reference destination is switched. ) To change the display mode of nodes and edges for the task corresponding to the record. The analysis / visualization program 195 changes the display mode of both the task behavior visualization tree and the data structure visualization tree as the designated time changes. That is, for example, when the state of task # 1 changes from an unstarted task to an OP completed task as a result of changing the designated time, the analysis / visualization program 195 displays the display mode of the node and edge corresponding to task # 1 Is changed from the display mode of the unstarted task to the display mode of the OP completed task (for example, the display mode in the middle is also changed by animation), and accordingly, the DB page # 101 referenced from the task # 1 is changed. The display mode is changed from the display mode of the non-IO page to the display mode of the OP completed page (for example, the display mode in the middle is also changed by animation). The analysis / visualization program 195 switches the record reference destination in the DB processing information 2400 in accordance with the change regardless of whether the designated time is changed in the forward direction (from the past to the future) or in the reverse direction (from the future to the past). The display of various trees can be changed.

  According to the second embodiment, the change in the display mode of the nodes and edges in the various trees drawn on the visualization window 2500 changes with the change in the designated time based on the DB processing information 2400. The system administrator can identify a problem that has occurred by looking at changes in the display mode of nodes and edges in various trees. For example, in the task behavior visualization tree, if the node and edge display mode transitions in the order of task IDs, the system administrator sequentially executes tasks even though the DBMS 120 can execute the tasks. It is possible to identify that there was a problem of execution. Further, for example, when there is a DB page whose display mode does not change in the data structure visualization tree, the system administrator has a problem that a task to access the DB page has not been executed. Can be identified.

  In the second embodiment, the analysis / visualization program 195, when an operation for zooming in is performed such as pressing the zoom-in button 2505, among the zoom-in targets, for example, the task behavior visualization tree and the data structure visualization tree. At least one can be enlarged and displayed. Further, when an operation for zooming out such as pressing a zoom out button 2506 is performed, the analysis / visualization program 195 displays at least one of zoom out targets, for example, a task behavior visualization tree and a data structure visualization tree. Can be displayed in a reduced size. In addition, various aspects can be employ | adopted as an aspect of such expansion / contraction. For example, a mode for expanding and contracting in both the vertical direction and the horizontal direction and a mode for expanding and contracting only one of them are provided, and the analysis / visualization program 195 is selected by the system administrator from those modes. You may expand and contract the tree in the mode. Further, as a mode, there may be a mode for adjusting the expansion / contraction rate so that the root node is always displayed. At the time of redrawing, the analysis / visualization program 195 may perform the drawing process only for the adjacent edge of the note included in the area displayed on the drawing coordinates.

  As a task behavior visualization tree displayed in the task display area 2501, a radical tree as shown in FIG. 30 may be employed. The analysis / visualization program 195 draws a radical tree as shown in FIG. 30 by arranging task nodes at the same level in a concentric circle centered on a certain task node (for example, a root task node). can do. The “same level task” may be a task having the same number of generations from a certain task, or a task accessing a page at the same distance from the root page in a certain data structure. Also for such a radical tree, for example, when the designated time changes from the first time “103.00” to the second time (for example, some time in the future after the first time), the analysis / visualization program 195 can change the display of the radical tree from the display shown in FIG. 30 to the display shown in FIG. 31 based on the DB processing information 2400.

  A radical tree may be employed as a modification of the data structure visualization tree.

  The vertical direction described above may be an example of a first direction, and the horizontal direction described above may be an example of a second direction (for example, a direction orthogonal to the first direction).

  In addition, at least one of task behavior and data structure (or database) visualization is displayed in a type other than the tree as described above, for example, A Tree Visualization Reference IEEE Computer Graphics and Applications, Vol. 31, No. 6. November 2011), pp. 11-15. Various visualizations may be employed.

Although several embodiments have been described above, the present invention is not limited to the above-described embodiments, and can be implemented in various modes. For example, the following analysis system can be derived based on the above description.
A database management system (DBMS) that dynamically generates a task for executing an operation in a query processing for a database stored in a storage device having a plurality of storage media and executes the dynamically generated task in parallel An analysis system for analyzing the behavior of
Storage resources,
A processor connected to the storage resource,
The DBMS generates event information that is a set of records including the state after the change of the task and the time at that time,
The processor is
Obtaining the event information from a computer executing the DBMS, storing the event information in the storage resource;
Based on the event information, a task object that is an object representing a task is drawn, and a display mode of the task object is set to a display mode according to a task state of the task object.
Analysis system.

  The analysis system refers to two or more records of all or part of the event information in time series, and changes the display mode of the task object to a display mode corresponding to the state after the task of the task object is changed. be able to. The “time series” mentioned here may be the oldest order (forward direction) in the record or the newest order (reverse direction) in the record.

  The event information may include information indicating which task is generated by which task. The analysis system may draw an edge representing a task generation relationship that connects the task object to the task object based on the event information.

100, 180 ... computer, 140, 142 ... network, 144 ... display, 150 ... storage device, 112 ... OS, 120 ... DBMS, 154 ... controller, 170 ... control program, 195 ... analysis / visualization program, 132 ... DB monitor information Output unit, 118... OS monitor information output unit, 174... ST monitor information output unit.

Claims (14)

An analysis system for analyzing the behavior of a database management system (DBMS) that dynamically generates and executes tasks for executing operations in a query processing for a database stored in a storage device having a plurality of storage media in parallel Because
Storage resources,
A processor connected to the storage resource,
The DBMS generates first information that is a set of records including the state of the entity and the time at that time,
The processor is
Obtaining the first information from a computer executing the DBMS; storing the first information in the storage resource;
Drawing an object representing an entity based on the first information, and setting the display mode of the object to a display mode according to the state of the entity represented by the object;
Analysis system. The entity is a task that the DBMS dynamically generates and executes in parallel.
The object is a task object representing a task.
The analysis system according to claim 1. The first information includes information indicating which task is generated by executing each task, and information indicating a generation order of one or a plurality of tasks generated by executing a certain task,
The task object is a node;
The processor draws a graph having edges between nodes representing the generation relationship and generation order of each task.
The analysis system according to claim 2. The processor is
The task object is drawn in a different display mode according to the task status,
For the task specified by the user, display detailed information on the record or page in the data structure constituting the database in the storage device and the information on the record or page accessed by the task.
The analysis system according to claim 2 or 3. The entity is a record or page in a data structure constituting the database in the storage device.
The analysis system according to any one of claims 1 to 4. The state of the entity includes a state where the record or page is accessed by a task.
The analysis system according to claim 5. The computer executing the DBMS stores second information representing a summary of the data structure of the database,
The record of the first information includes the ID of the page to which the task corresponding to the record is accessed;
The processor is
Obtaining the second information from the computer; storing the second information in the storage resource;
Based on the second information, a page object, which is an object representing a page in the data structure constituting the database in the storage device, is drawn, and the display mode of the page object is a task for accessing the page. To display according to the state,
The analysis system according to claim 5 or 6. The page object is a node;
The processor draws a graph having an edge between nodes representing a dependency relationship between pages for each data structure of the database.
The analysis system according to claim 7. The processor is
The page object is drawn in different display modes depending on the state of access to the page,
Displays detailed information about the page's logical and physical addresses for the page specified by the user.
The analysis system according to claim 7 or 8. The processor changes the display mode of the object over time according to a change in task state associated with referring to two or more records of all or part of the first information in time series.
The analysis system according to any one of claims 1 to 9. 11. The processor according to claim 1, wherein the processor can switch between a display form for displaying all of the objects and a display form for displaying a part of the object in an enlarged manner according to a user instruction. The analysis system described in the section. A computer that executes a database management system (DBMS) that dynamically generates and executes a task for executing an operation in a query processing for a database stored in a storage device having a plurality of storage media in parallel;
An analysis system for analyzing the behavior of the DBMS;
The DBMS generates information that is a set of records including the state of the entity and the time at that time,
The analysis system includes:
Obtaining the information from a computer executing the DBMS;
Based on the information, the object representing the entity is drawn, and the display mode of the object is changed to the display mode according to the state of the entity represented by the object.
Computer system. Analysis method for analyzing the behavior of a database management system (DBMS) that dynamically generates and executes tasks for executing operations in a query processing for a database stored in a storage device having a plurality of storage media in parallel Because
Get information that is a set of records including the state of the entity and the time at that time,
Based on the information, the object representing the entity is drawn, and the display mode of the object is changed to the display mode according to the state of the entity represented by the object.
analysis method. For visualization of the behavior of a database management system (DBMS) that dynamically generates and executes tasks for executing operations in the processing of a query for a database stored in a storage device having a plurality of storage media in parallel A computer program,
Get information that is a set of records including the state of the entity and the time at that time,
Based on the information, the object representing the entity is drawn, and the display mode of the object is changed to the display mode according to the state of the entity represented by the object.
A computer program that causes a computer to execute.