JPH09330302A - Method and system for monitoring performance of computer system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable more than one computer to monitor performance data on a computer to be monitored object through a network without increasing the load on the object computer. SOLUTION: A sampling process 4 actuated on each node 2 of parallel computers 1 samples performance data and a gathering process 3 actuated on a specific node gathers those performance data and sends them to a receiving process 15 on a monitor computer 11. The receiving process 15 distributes the performance data to a display process 16 or storing process 17 which is already actuated on the same or different monitor computer 11 when such a process is present. The display process 16 displays some of itemized performance data included in the performance data on a display device 12. The storing process 17 stores all the performance data in a storage device 13.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a performance monitoring method for collecting and displaying performance data in a computer system in which a plurality of computers are connected via a network, a computer system and a program storage medium therefor, and particularly a parallel computer or The present invention relates to a performance monitoring method suitable for a distributed system.

[0002]

2. Description of the Related Art In a parallel computer or a distributed system, a plurality of computers, which are called nodes constituting the parallel computers, operate in parallel in cooperation with each other. Depends on the behavior of the node. Therefore, the operation becomes very complicated as compared with the sequential computer.

In order to effectively use such a parallel computer and bring out sufficient performance, not only the operation of a single node but also a complicated operating condition including a causal relation of the operation between the nodes and a load balance, etc. It is necessary to accurately grasp the information and use the information for tuning the running program.

The following two methods have been mainly used as conventional techniques for supporting the grasp of the operating status of a computer.
The first is the method used in PerfView sold by Hewlett-Packard, etc., and the operating status of the CPU such as CPU operating status, memory usage status and network communication frequency for each node of the distributed system. Performance data regarding the situation is measured and stored in a storage device included in the node, such as a magnetic disk storage device. The performance data accumulated in each node is accumulated in one computer connected to the distributed system, and displayed so that it can be visually understood by graph display.

Secondly, IBM's Visualizat
In the method represented by ion Tool, a process for collecting performance data is started on each node of the parallel computer, and the display process started on the control computer connected to the parallel computer via the network is Performance data is received and displayed in real time from the process on the node. For example, US International B
usins Machines Corp. Published by "IBM Parallel Environneme
nt for AIX Operation and
Use Version 2.1.0, "pp.263.
-265, 1995 (Document number GC23-3891-0
See 0).

[0006]

Generally, a parallel computer or a distributed system is commonly used by a plurality of users. Therefore, it is desirable that a plurality of users can monitor such computer performance data in real time via a network.

In the first method, however, the performance data can be stored in the storage device for each node, but since the performance data measurement is once completed and the data is analyzed and displayed, the operating status is real-time. I can't grasp.

In the second method, the performance data can be received and displayed in real time from the process on each node, but in the above reference, a plurality of users monitor the performance data of the same monitored computer. The method is not specifically shown.

When a plurality of users monitor the performance data of the same monitoring target computer, it is further desirable that the monitoring load on the monitoring target computer does not increase as the number of users increases. .

Accordingly, an object of the present invention is to provide a computer performance monitoring method suitable for a plurality of users to monitor performance data in real time while suppressing an increase in the load of the computer whose performance is to be measured, and a computer system therefor. And providing a program storage medium.

[0011]

To achieve the above object, in the present invention, performance data is repeatedly collected by a computer to be monitored and transmitted to a computer for operating status monitoring via a network. Invokes a receiving process on this monitor computer, and this process receives the transmitted performance data and receives it on one or more utilization processes started on multiple computers that are the same as or different from that computer. Distribute the performance data. The measured performance data includes a plurality of performance data for a plurality of measurement items. The utilization processes are, for example, a plurality of display processes or a plurality of accumulation processes.

When any of the display processes receives the performance data transferred by the receiving process, the display process displays the performance data for some measurement items among the transferred performance data. Display on the display device connected to the running computer. When any of the storage processes receives the performance data transferred by the reception process, the storage process stores all the transferred performance data in the storage device connected to the computer on which the storage process is activated. To do.

In one more preferable operation mode of the present invention, when the computer to be monitored is a parallel computer, the performance data collected by the collection process activated on each node of the parallel computer is connected to the network. Was done,
The data is collected by a collection process activated on a predetermined one node in the parallel computer and transferred to the reception process.

[0014]

BEST MODE FOR CARRYING OUT THE INVENTION The performance monitoring method according to the present invention will be described below in more detail with reference to some embodiments shown in the drawings. In the following, the same reference numerals represent the same or similar ones. Further, in the second and subsequent embodiments, differences from the first embodiment will be mainly described.

<First Embodiment of the Invention> In FIG.
The parallel computer 1 includes a plurality of nodes 2 each including at least one processor 2A and memory 2B,
The internal network 5 connects between these nodes. The memory 2B of each node holds a program and data executed by the processor 2A of that node.
Each node or some of the nodes further has a peripheral device such as a magnetic storage device, but they are not shown here for simplification. One specific node of the parallel computer 1 is connected to the external network 21, and the external network 2
A plurality of computers 11 are connected to 1, and the above-mentioned one node and these computers can communicate with each other via this network 21. The parallel computer 1 is a computer to be monitored, and two computers 11 connected to the external network 21 are examples of computers used for monitoring. Other computers connected to the network 21
Not shown for simplicity. Each computer 11 includes a processor 11A and a memory 11B, and the computer 11 is connected to an input / output device 12 including a display device and a keyboard and a storage device 13 such as a magnetic disk storage device. Each processor has an appropriate O
S, eg X / Open Company Lim
UNI developed and licensed by ited
Controlled by X.

FIG. 1 shows the five types of processes that make up the monitoring system of this embodiment and the relationships between them. These processes are ones in which five programs recorded in an appropriate program recording medium are installed in the parallel computer 1 and executed as respective processes. Each node 2 of the parallel computer 1 has a collection process 4. The collection process 4 has a function of repeatedly collecting the performance data of each node 2 at regular time intervals. The collection process 3 is placed on a specific node connected to the external network 21 among the plurality of nodes 2 of the parallel computer 1, and the collection process 4 on each node
The performance data of each node collected by is collected and transmitted to one of the monitoring computers 11.

On the other hand, a reception process 15, a display process 16 and a storage process 17 are activated in this one monitor computer 11. The reception process 15 is activated by only one of the plurality of monitor computers, and transmits / receives data to / from the collection process 3 on a one-to-one basis. The display process 16 and the accumulation process 17 are activated by a required number of users or users, and receive performance data from the reception process 15. The display process 16 displays some item-specific performance data in the performance data on the display device in the input / output device 12. The storage process stores the entire performance data in the storage device 13.
The display process 16 and the storage process 17 are not necessarily the monitor computer 11 in which the reception process 15 is activated.
It is not necessary to be started on the above, and those processes can be started on another computer connected to the monitor computer by the network 21. In addition, a plurality of display processes 16 are provided on any one monitor computer 11.
Can also be started. Similarly, it is possible to activate a plurality of storage processes 17 on the computer.

Next, how each process of this monitoring system operates in cooperation with each other will be described with reference to FIG.
Flowcharts of (a) and (b) and FIGS. 2 to 5
Will be described with reference to the internal configuration diagrams of the above five processes shown in FIG.

First, the collection process 3 is started on the specific node 2 connected to the external network 21 on the parallel computer 1 (step 521 (FIG. 6A)). In normal operation, the collection process 3 is started by the system administrator issuing a collection process start command to the parallel computer 1. When started, the collection process 3 first executes an initialization process. For example, the configuration definition file 208 (FIG. 3) in which the configuration of the parallel computer 1 is described, such as the number of nodes, the attributes of each node, and the number of nodes that acquire performance data, is read. In this embodiment, it is assumed that all nodes acquire performance data. The attributes of each node include information such as the address of the node, the peripheral device attachment status indicating whether or not a peripheral device such as a magnetic disk storage device is attached to the node. When the initialization process is completed, this collection process 3 will be activated on any one of the monitor computers 11
Wait for connection request from 5.

Next, any one monitor computer 1
1 to start the receiving process 15 (step 54)
1). In normal operation, the reception process 15 is activated by the system administrator issuing a reception process activation command from the monitoring computer 11. This receiving process 15 can be activated on any computer connected to the parallel computer 1 by a network.
However, since this process may be connected from multiple display processes started by multiple users on different computers, it is possible to specify a specific computer that can be accessed by multiple users, such as a parallel computer management workstation. It is generally started on a computer. The system administrator uses the IP of the parallel computer 1 that is required when issuing a connection request to the collection process 3 as an argument of the start command of the reception process 15.
Input parameters such as address information (or host name) and time interval for collecting performance data.

The activated receiving process 15 makes a connection request to the collection process 3 by the connection processing routine 305 of FIG. 4 (step 542). This connection request is U
It can be implemented by the connect system call in NIX. At this time, the IP address information of the parallel computer given as an argument at the time of start-up is used as the identification information of the collection process 3 which is the other party who issues the connection request. In the collection process 3 that has received the connection request, the connection processing routine 202 performs connection processing with the connection processing routine 305 (FIG. 4) in the reception process 15, whereby the collection process 3 communicates with the reception process 15. Allows data to be sent and received (step 522).

After the connection process is completed, the collection process 3 sends the configuration definition information of the parallel computer 1 previously read from the configuration definition file 208 to the reception process 15 (step 523). Now, in this system, requests are transmitted and data is transmitted and received by exchanging messages between the processes that configure the system. In the present embodiment, the message is a variable-length byte string, the first byte is an identifier field that stores an identifier indicating the type of message, and data is stored in the subsequent data field. The message may consist only of the identifier field without the data field. Step 52 above
3, the internal processing routine 207 (FIG. 3) in the collection process 3 is provided for the reception process 15 in order to transmit the above configuration definition information.
A message including the configuration definition information is constructed on the 03. That is, a 1-byte identifier indicating that this message includes the configuration definition information is stored in the identifier field,
The IP address of the monitoring computer 11 is stored in the data field. The message thus configured is sent to the receiving process 15 via the network 21 by using the UNIX send system call.

Upon reception of this message from the collection process 3, the reception process 15 receives the input analysis routine 306.
In FIG. 4, it is detected from the identifier that the message content is the configuration definition information of the parallel computer 1. Further, the sizes of the input / output buffers 301 and 302 required for transfer of performance data are calculated from the number of nodes included in the configuration definition information. The calculated size is used later when those I / O buffers are reserved. Each input / output buffer 30
1 or 302 is reserved corresponding to one of the display process 16 or the accumulation process 17 which is activated later.
The size of these input / output buffers may be at least as long as the length of the performance data collected by one collection process 4 in one hour step is multiplied by the number of nodes in the parallel computer. Next, the reception process 15 requests the collection process 3 to activate the collection process 4 of each node (step 543). Specifically, similarly to the above-mentioned exchange of computer configuration definition information, the collection process 4 sends a message including a request identifier indicating that it is a start request to the collection process 3, and the collection process 3 that has received the message sends the message. The input analysis routine 205 identifies the content of the request from the identifier.

The collection process 3 requested by the receiving process 15 to start the collection process 4 starts the collection process 4 on each node 2 using the node address in the previously received configuration definition information (step 524). The remote shell function of UNIX is used for activation. When the collection process 4 is started (step 501), UNIX c
Initial processing, such as executing connection processing with the collection process 3 by the connection processing routine 100 (FIG. 2) using the connect system call, waits for the monitoring start request from the collection process 3. In this way, when the collection process 4 is activated and the connection process with the collection process 3 by the collection process is completed, the collection process 3 notifies the reception process 15 of the completion of the activation process. Receiving it, the receiving process 15 requests the collection process 3 to start monitoring (step 544). At this time, the message sent by the receiving process for the monitoring start request includes the identifier indicating the content of the request and the information about the performance data collection time interval.

Upon receiving the monitoring start request, the collection process 3 transfers the start request message to the corresponding collection process 4 (step 525). When the collection process 4 is activated, the counter control / readout routine 104
The select system call is issued, and the collection process 4 enters the message arrival waiting state. The sampling process 4 is s
If a message arrival is detected by the value of the return to the elect system call, this collection process 4
The process moves to the input analysis routine 102. The input analysis routine 102 issues a receive system call to read a message into the input / output buffer 101. It is confirmed that the identifier at the beginning of the message is the monitoring start request identifier, the performance data collection time interval included in the message is extracted, and this is used as a return value to return to the counter control / readout routine 104. The collection process 4 receiving the collection request in this way reads the collection time interval included in the request, and the counter control / readout routine 104 sets this collection time interval in the clock 107, and the clock 107 sets the set time interval. Each time an interrupt is generated in the counter control / readout routine 104.

Counter control / readout routine 104
Whenever it receives an interrupt from the clock, U
A function such as rstat, which is a system call for collecting performance data of NIX, is issued to collect performance data from the OS 105 (step 502). The performance data is composed of a plurality of item-specific performance data, and the number and types of item-specific performance data that can be collected are predetermined according to the specifications of the above functions. Generally, it is possible to obtain item-wise performance data such as CPU utilization rate, memory utilization rate, magnetic disk storage device access frequency, network communication frequency (that is, transmission frequency and reception frequency) within a unit time. The OS reads these performance data from the software counter 106 in the kernel or the hardware counter 108 in the node and returns it as an output argument of rstat.

The counter control / readout routine 104 of the collection process 4 that collects the performance data in this way
Stores the performance data in the input / output buffer 101,
The control is transferred to the output control routine 103. The output control routine 103 issues a send system call, which causes the contents of the input / output buffer 101 to be transferred to the internal network 5
To the collection process 3 via (step 50).
3). In the collection process 3, one of the collection processes 4
The internal processing routine 207 (FIG. 3) monitors the arrival of a message including performance data from the OS using a select function which is a system call of the OS. Upon confirming the arrival of the message, the internal processing routine 207 activates the input analysis routine 205. The input analysis routine 205 is
A receive system call is issued to read the performance data in the message into the input / output buffer 201. When the reading of the performance data is completed, the input analysis routine 205 stores the collection process of the transmission source, confirms whether the performance data has been transmitted from all the collection processes 4, and returns the processing to the internal processing routine 207.

At this time, if there is a collecting process 4 that has not yet transmitted the performance data, the internal processing routine 20
7 issues a select system call to continue monitoring the arrival of messages containing performance data. When the performance data from all the collection processes 4 arrive, the internal processing routine 207 receives the performance data from the collection processes 4 from the input / output buffer 201 provided corresponding to each collection process, and receives the performance data. The data is copied to the I / O buffer 203 for 15, and the performance data on each of the copied nodes is assembled into one message. The message is composed of performance data in which the performance data sent from each collection process is linked after the identifier. Next, the internal processing routine 207 activates the output control routine 206. Output control routine 206
Issues a UNIX send system call and sends a message containing this performance data to the receiving process 15 (step 526).

In the receiving process 15, the internal processing routine 308 issues a UNIX select system call and waits for the arrival of a message from the collecting process 3. Upon detecting the arrival of a message from the collection process 3, the internal processing routine 308 shifts the processing to the input analysis routine 306. The input analysis routine 306 is a UNIX rec
An eive system call is issued to read the performance data in the message into the input / output buffer 304 provided corresponding to the collection process 3 (step 545). After receiving the performance data, the reception process 15 confirms whether the display process 16 or the storage process 17 is connected. If no display process is connected, the following performance data transfer processing is not performed. When at least one display process 16 is connected, the receiving process 15 copies the message arriving at the input / output buffer 304 to the input / output buffer 301 provided on a one-to-one basis with respect to the display process 16. Thereby, the message is distributed to the display process 16 connected to the input / output buffer 301. If there is an input / output buffer 301 to which a plurality of display processes 16 are connected, the performance data is distributed to all the display processes 16 in exactly the same manner. When the storage process 17 is connected, the input / output buffer 301 provided in connection with the storage process is used. Details of this distribution processing will be described later.

In this way, after the collection process 4, the collection process 3, and the reception process 15 are activated by the system administrator, the user of this system displays the display process 16.
Alternatively, if the storage process 17 is activated, it becomes possible to monitor the performance data using these processes. In this case, the other monitor computer 11 shown may be a desk computer used by one user of the parallel computer 1. In this way, the performance data of the parallel computer 1 can be monitored from the computer 11 managed by the system administrator or from a general-purpose computer 11 for personal use, which is remote from the computer and is managed by the user. In this specification, even a general-purpose computer that is not exclusively used for performance data monitoring is referred to as a monitor computer even if it is used for performance data monitoring.

In this embodiment, the receiving process can be activated on any computer, and the display process or the storage process using the performance data can be activated on any computer. Further, a plurality of predetermined item-specific performance data are collected without depending on the item-specific performance data required by each display process, and the display process selects and uses the item-specific performance data from among them. If the number of these predetermined plural items is sufficiently large, the user's request can be usually satisfied by the above method even when the user requests other item-specific performance data.

Further, in this embodiment, all the collected item-specific performance data are stored in the storage device in the accumulation process so that the user can select the item-specific performance data to be used later. There is. For this reason, a predetermined number of item-specific performance data are collected regardless of what the item-specific performance data requested by the user is.

Furthermore, the collection process, the collection process, and the reception process are activated independently of the activation of the display process or the accumulation process, and the performance data collected in response to the activation of the display process or the accumulation process thereafter is sent to those processes. We are distributing. As a result, the reception process, the collection process, and the collection process only have to execute the same processing even if the presence or the number of the activated display processes changes.

The operation of the display process 16 is as follows. The display process 16 is activated on the monitor computer 11 used by the user of the monitoring system (step 561 (FIG. 6B)). At this time, the user confirms in advance the IP address of the computer on which the receiving process 15 is activated and designates it as an argument when the display process 16 is activated.

When a plurality of users use this monitoring system, a different monitor computer is usually used for each user. For example, two users use each of the two monitor computers 11 shown in FIG. In this embodiment, the same user can activate a plurality of display processes on the same computer 11. At least one display process (first type display process), as exemplified later, displays the item-specific performance data defined for the display process in the form of a graphic or graph defined by the display process. Is configured. Other display processes (type 2 display process)
Is configured to display the performance data for each item selected by the user by a graph predetermined for the display process. As the latter kind of display process,
A plurality of display processes prepared corresponding to different graphs can be activated on the same monitor computer. Further, both types of display processes are configured to request the transfer of performance data to the receiving process according to a user's instruction after being activated. As described above, since the display process is configured to be activated for each graph that can be displayed, the structure of each display process is simple.

As a result, in the present embodiment, in order to display the performance data on the same computer, the user needs to start at least one display process. Since the user desires to display the performance data, the same user starts a plurality of second type display processes on the same computer. At that time, a second type display process to be activated is selected according to which graph the performance data is displayed, and after the selected second type display process is activated, the display process should be displayed. It is designed to indicate performance data by item.

In the activated display process 16, the control routine 403 performs initialization processing for initializing the display screen and initialization for connecting to the receiving process 15, and then shifts the processing to the connection processing routine 406 to the receiving process 15. A connection request is issued (step 562).

Upon receiving the connection request, the connection processing routine 303 of the receiving process 15 generates the input / output buffer 301 and the request flag 302 for the display process,
Connection processing is performed for the display process (step 5).
46). The connection process for the above receiving process is completed,
When data can be transmitted / received to / from the process, the reception process 15 transmits the configuration definition information of the parallel computer 1, which the reception process 15 has previously received from the collection process 3, to the display process 16 that has just been connected. To do.

Display process 16 that received the configuration definition information
Displays the window on the display device after securing the transmission / reception buffer 404 for accumulating the performance data by referring to the information and calculating the graph layout for displaying the performance data of all the nodes. Yes (Step 56)
3). FIG. 7 shows a display process (first type display process).
5 is an example of a window that displays a plurality of item-specific performance data predetermined by the above. The operating status of one node is represented by two rectangular parallelepipeds and a pair of arrows (601, 602, 603). By arranging a plurality of such figures corresponding to a plurality of nodes, the operation status of the entire parallel computer is represented. The heights of the two rectangular parallelepipeds 601 and 602 respectively correspond to the CPU utilization rate and the memory utilization rate in one node, and the length of a pair of arrows 603 corresponds to the number of transmissions and the number of transmissions and receptions from the node to the network 21. Correspond. The display window 610 includes a display area 620 and a control area 630, and the user controls the display process by using objects such as buttons 640 and 650 arranged in the control area 630. The data display is started by operating the button 640, and the display is stopped by operating the button 650.

FIG. 8 is an example of a window displayed by another display process (second type display process) for displaying one type of item-specific performance data selected by the user on one screen. A bar graph is displayed in the display area 620. The horizontal axis 1711 of the graph corresponds to different nodes, and the vertical axis 171
One item-specific performance data is mapped to 2. The user selects the performance data to be mapped on the vertical axis of the bar graph from the performance data list 1704 provided in the control area 630. Buttons 640 and 650 are similar to those in FIG. In either case of the display process of FIG. 7 or FIG. 8, when the user instructs the start of data display by operating the button 640 in the display window 610 or 1700, the display process 16 receives the reception process 15
A performance data transfer request message is transmitted to (step 564).

As described above, the receiving process 15 issues a select system call in the internal processing routine 308, and is in a state of waiting for a message arrival in the input / output buffers 301 and 304. Select by the arrival of the data transfer request message from the display process 16.
The system call returns, and the internal processing routine 308
When the arrival of this message is detected, the process shifts to the input analysis routine 306. The input analysis routine 306 is r
An input / output buffer 301 connected to the display process by issuing the receive system call to send a request message.
Read in. Next, the output control routine 307 checks the identifier of the request message, determines that the message is a data transfer request, and outputs the I / O buffer 30
The request flag 302 associated with 1 is set. In addition, when a plurality of display processes 16 or accumulation processes 17 are activated by the same user or different users,
The reception process 15 repeats the above processing and performs connection operation with all of these processes.

The operation of the apparatus after the display process 16 is thus connected to the reception process 15 and the display is started by the screen operation of the user will be described. The performance data is transferred from the collection process 4 to the reception process 15 via the collection process 3 by the above-described procedure (steps 504, 527, 548). At this time, the receiving process 1
5 issues a select system call in the internal processing routine 308, and is in a state of waiting for a message to arrive at the input / output buffers 301 and 305. Se is triggered by the arrival of a message containing performance data from the collection process 3.
The lect system call returns to the internal processing routine 308, and the internal processing routine 308 transfers control to the input analysis routine 306. The input analysis routine 306 issues a receive system call to read the arrived message into the input / output buffer 304, checks the message identifier, confirms that this is a message including performance data, and processes it in the internal processing routine 308. return. The internal processing routine 308 is the display process 16
The request flags 302 associated with the input / output buffers 301 connected to are sequentially checked (step 547).

If the request flag 302 associated with any one of the input / output buffers 301 is set, a message including performance data is copied from the input / output buffer 304 to one of the input / output buffers 301, and the output control routine 307 is executed. To start. The output control routine 307 issues a send system call to transfer the above message stored in the I / O buffer to the display process connected to the I / O buffer 301, and clears the request flag 302 attached to the I / O buffer 301. (Step 549). This processing is performed for all request flags 3 provided for all connected display processes 16.
The reception process 15 distributes the performance data to all the display processes 16 by repeating the process for 02. The same applies when the storage process is connected to the input / output buffer 301. Since the receiving process 15 on the monitor computer 11 performs the processing of distributing the performance data to the plurality of display processes 16 as described above, even if the number of activated display processes 16 is increased, The collection process 4 and the collection process 3 in 1 above are not affected, and the load on the parallel computer to be monitored does not increase.

The display process 16, which has received the performance data transferred from the reception process 15, creates a display screen by referring to the item-specific performance data necessary for display out of all the received data, and draws it on the input / output device 12. Yes (Step 56)
6). In the case of a display process having the window shown in FIG. 7, the data necessary for display is the CPU utilization rate, the memory utilization rate, and the number of communications, which are defined by the display process.
It is one item-specific performance data, and in the case of the display process having the window of FIG. 8, it is one item-specific performance data instructed by the user for the display process.

The display process 16 displays the performance data sent one after another by repeating the display operation from processing 564 to processing 566. That is, the control routine 403 of the display process 16 sends a data transfer request to the receiving process 15 (step 564).
The specific operation at that time is as described later.
The control routine 403 then waits to detect either the user's windowing or the arrival of a message containing performance data from the receiving process. When the message including the performance data arrives, the control routine 403 issues a receive system call to read the message into the transmission / reception buffer 404 (step 565) and shifts the processing to the input analysis routine 401.
The input analysis routine 401 confirms that the identifier of the message sent from the receiving process 15 is the identifier of the message including the performance data. Next, the drawing processing routine 402 is activated, and the drawing processing routine 402 changes the height of the graph on the window according to the performance data (step 566). When the drawing is completed, the process returns to the control routine 403, and the control routine 403 creates a data transfer request message for the performance data of the next time step in the transmission / reception buffer 404, issues the send system call, and receives this message in the receiving process 15 (Step 564). After sending, control routine 4
03 returns to the waiting state again.

On the other hand, the stop of the display is realized by stopping the transmission of the performance data transfer request message which is being transmitted after drawing the performance data for each time step. When the user operates the button 605 in the display window 610, the control routine 403 exits the waiting state and transfers control to the input analysis routine 401. The input analysis routine 401 analyzes the request and informs the control routine 403 that the pressed button is the stop button 650. When the control routine 403 detects this state,
After that, the performance data transfer request message is not transmitted to the receiving process 15. The display process 16 that does not make a data transfer request does not set the corresponding request flag 302. Therefore, the receiving process 15 does not send performance data to such a displaying process 16. Therefore, the display of the display process 16 is stopped.

The ending process of the display process 16 is executed by the menu operation of the display window 610. Similarly to the case of the stop processing, also in the case of the end processing, the control routine 403 detects the user's window operation, exits the waiting state, and shifts the processing to the input analysis routine 401. The input analysis routine 401 detects that stop has been selected from the menu and notifies the control routine 403. Control routine 40
3 forms an end notification message in the transmission / reception buffer 404 and sends it to the receiving process 15 by the send system call (step 567).

The receiving process 15 receiving the end notification message cancels the connection state with the display process 16,
The allocated input / output buffer and request flag are released (step 550). When the connection is released, the display process 16 performs termination processing (step 568) and then terminates (step 569). When the connection with all the display processes 16 is released, the system administrator can start the termination process of the receiving process 15.
To start the termination process of the receiving process 15, an interrupt signal is input to the receiving process 15 by using the interrupt function provided by the OS that controls the computer 11 on which the receiving process 15 is activated. The receiving process 15 that has received the interrupt signal issues an end request to the collecting process 3 (step 551). The collection process 3 that has received the termination request issues a termination request to all the collection processes 4 (step 528), and the collection process 4 that has received the termination request performs termination processing (step 505). When the connection to all the collection processes 4 is released, the collection process 3 releases the connection to the reception process 15. After that, the collection process 3 and the reception process 15 each independently perform termination processing, and all the processes in the monitoring system are terminated (steps 529 and 552).

Next, a method in which the accumulation process 17 accumulates the performance data in the storage device 13 and the accumulated data is displayed by the display process 16 will be described. The accumulation process 17 shown in FIG. 5 is activated by the user similarly to the display process 16. When the storage process 17 is activated, the connection processing routine 455 issues a connection request to the reception process 15 and connects to the reception process 15 in the same procedure as that executed by the display process 16.
A control window (not shown) for allowing the user to control the operation of 7 is displayed. The control window is similar to the display window of display process 16, eg, 610 (FIG. 7), but without the display area 620.
It consists of only 30. In the control area 630, the user can control selection of a data item to be stored, designation of a file name to be stored, start and end of storage. When the data accumulation start operation is performed after setting the data items to be accumulated and the file name to be accumulated by the operation on the control area, the accumulation process 17 receives the performance data similarly to the display process 16. The input analysis routine 451 confirms that the message identifier sent from the reception process 15 is the message identifier including the performance data. This performance data is stored in the transmission / reception buffer 452, and thereafter, this performance data is stored in the data shaping routine 4
Output routine 4 is formatted in 53 for storage data format
A storage device 13 such as a magnetic disk storage device by 54
Is accumulated in.

The format of the file 1004 to be stored is shown in FIG.
As shown in (d), first, the configuration definition information of the parallel computer such as the total number of nodes of the parallel computer, the attributes of each node described above, the list of nodes from which data is collected, etc. is stored. After that information, multiple blocks are stored. Each block corresponds to one of the plurality of nodes and one of the plurality of time steps. Each block consists of multiple data records, each data record containing performance data for different measurements taken from the same node at the same time step. The header record 1001 of FIG. 9A and the header record 1001 of FIG.
A block end record 1003 of (b) is placed and indicates a block boundary. In the header record, the block length that indicates the length of the entire block, the number of the node that generated multiple data records belonging to that block, and the time information that indicates the time step at which the performance data included in those data records was acquired, etc. It is included. Further, at the beginning of every record, there is a type code indicating the type of record and a record length indicating the length of the record. The order in which the blocks are arranged is such that one block corresponding to a specific node and a specific time step is stored first, and then a plurality of other nodes for other nodes for the same time step are stored. After accumulating multiple blocks for all nodes for the same time step,
The blocks are arranged for the subsequent time steps.

The data accumulated by the accumulation process 17 in this way can be displayed by any of the display processes 16. The display process used for this display may be a display process that has already been started and is displaying performance data in the display window shown in FIG. 7 or FIG. 8, or a display process that has not been started may be used. In the following, the case of using the display process that has already been started and is used for displaying the performance data will be described.

The flowchart of FIG. 10 shows the display process 1.
6 shows the procedure for reading the accumulated data and displaying it. First, the user inputs the file name in which the accumulated data to be displayed is stored from the input / output device 12 in advance (block 1101). After that, when the user inputs a request for switching to the accumulated data display mode to the display process 16, the display process 16 switches to the accumulated data display mode (step 1102). At this time, the display process 16 stops transmitting the data transfer request to the receiving process, and the input destination switching routine 405 (FIG. 5) switches the data input destination from the receiving process 15 to the designated file on the storage device 13.

When the display process 16 opens this file (step 1104), the display process 16 reads the configuration definition information of the parallel computer at the head of the file (step 1105). At this time, the configuration definition information used up to that point is saved to a buffer (not shown), the layout of the graph is recalculated based on the configuration definition information read from the file, and the graph is displayed on the display screen. To do. Next, the file is read to the end, the number of blocks is counted, and this number of blocks is divided by the number of nodes in the configuration definition information read earlier to calculate the number of time steps at which multiple blocks are measured. (Step 1
106). Then, the control screen of the accumulated data display function is displayed (step 1107).

FIG. 11 shows an example of the control screen. The file name display column 902 displays the name of the file currently being read, and the data number display column 903 displays the number of time steps of accumulated data calculated in the processing block 1106. In the display range input fields 904, 905, and 906, the number of time steps to skip, such as the beginning, the end, and at what time step of the time step range to be displayed, is input. Slider 907
Indicates the number of the time step at which the data is currently displayed, and the display data can be changed by moving this position. Buttons 908, 9
Reference numerals 09, 910, and 911 are buttons for designating frame advance to the previous time step, stop of continuous display, start of continuous display, and frame advance to the next time step, respectively.

At step 1109, when the user designates the display time, that is, the time step of the data to be displayed, on the control screen 901, the display process 16
Reads the contents of the file and retrieves the data of the corresponding time step (step 1110). Then, the data is drawn on the display device (step 1111).
Further, in step 1109, the button 9 on the control screen 901 is pressed.
When the continuous display is started by pressing 10, the current time until the last time step of the display range specified in the display range input field 905 is reached or until the continuous display is stopped by the display button 909 (step 1112) Data for one hour step is read from the next to the read data (step 1113) and displayed repeatedly (step 1114). The time interval for repeating the display may be calculated from the time information in the header record 1001 accumulated in the file or may be specified in advance. In any case, the display process 1 in FIG.
The clock generation routine 407 in 6 measures the time, and when the designated time has elapsed, the drawing process is performed.

When the user instructs the end of the accumulated data display function from the input device such as a mouse (step 11).
08), end processing is executed, whereby the input destination switching routine 405 switches the input destination to the receiving process, and the display process 16 returns to the normal data display mode (step 1115). At this time, the display process 16 discards the configuration definition information read from the file, restores the previous configuration definition information saved in the buffer,
Restores the graph in the display area to the real-time display status before switching to the accumulated data display mode. As a result, by starting the accumulation process 17 and the display process 16, it becomes possible to redisplay the past data while the current data is being displayed in real time. It is also possible to perform post-mortem analysis of the operating status of the parallel computer using the data accumulated in advance. The accumulation process 17 is the computer 1 for each monitor.
The performance data can be stored in parallel in the storage devices 13 included in the computers that have been started up in parallel and that have been started by the respective storage processes.

Furthermore, since the storage process 17 is not particularly distinguished from the display process 16 from the viewpoint of the reception process 15, the fact that a plurality of display processes 16 can be activated for one reception process 15 means that This means that a plurality of storage processes 17 can be activated. Therefore, a plurality of users can simultaneously use different data storage functions by simultaneously activating different storage processes 17, and the load on the parallel computer does not increase due to the activation of these storage processes.

<Second Embodiment of the Invention> The present invention can be applied to a computer different from the parallel computer used in the above embodiments. In the above embodiment, only a specific node in the parallel computer is connected to the external network.
However, some of the parallel computers already developed have a plurality of specific nodes connected to this external network. In such a parallel computer, a collection process is placed in each of those specific nodes,
In the parallel computer, the collection of performance data and the transfer to the receiving process can be divided and executed by the plurality of collecting processes.

That is, using the communication between the receiving process and the collecting process activated in each specific node, the collecting process collects performance data of some nodes including the specific node. The collecting process collects the performance data collected by those nodes and transfers it to the receiving process. This is done in parallel with each other performed by the collection process launched on each particular node. By doing so, the load of each specific node in the second embodiment is reduced as compared with the load of one specific node collecting a plurality of performance data collected by all the nodes in the previous embodiment. . Even in such a parallel computer, even when a plurality of users try to monitor performance data in the parallel computer, the load on the parallel computer does not increase.

<Third Embodiment of the Invention> Another parallel computer already developed is divided into a plurality of partitions each including a plurality of nodes. In this case, for each partition, any one of the jobs can be executed in parallel by a plurality of nodes in the partition. Of course, it is also possible to execute a plurality of jobs in each partition. However, it is suitable for one user to occupy and use one partition. In such a computer system, it is effective to monitor the operating status for each partition.

For example, in FIG. 13, the parallel computer 1 has two partitions 1 and 2 (100A, 1A).
00B). A predetermined one node of the plurality of nodes included in each partition is connected to the external network 21. In this case, the collection process 4 is started on each node, and the collection process 3
Is started on a specific node 2 of each partition connected to the external network 21. In order to monitor the performance data of the partition 1, the first receiving process 15
A is started up on the first monitor computer 11 to communicate with the collection process 3 started up on a specific node of the first partition. Further, the display process 16A or the storage process 17A is activated on the monitor computer 11 and is connected to the first receiving process and 15A. Similar to the case of the first embodiment, the display process 16A or the storage process 17A receives the performance data of the partition 1 from the receiving process 15A. When another display process or storage process (not shown) is activated by another computer (not shown), the display process or storage process also receives performance data distribution via the first receiving process.

Similarly, for the other partition 2, the second receiving process 15B, the display process 16B or the storage process 17B, or another display process or storage process (not shown) is activated. In this embodiment, performance data can be collected for each partition independently of other partitions. Furthermore, even if both the display process and the accumulation process that monitor the performance data of the same partition are launched on the same monitor computer, or if multiple display processes that monitor the performance data of the same partition are used for different monitors. Even when started on the p computer, the load on the parallel computer does not increase.

<Fourth Embodiment of the Invention> The most extreme case of the second embodiment is a case where all the nodes of the parallel computer are connected to the external network. In this case,
It suffices that each node activates the collection process and the process having the functions of the collection process, and the reception process communicates with this process of each node to receive the performance data of that node. In this case, since the same process for collecting performance data is started in each node, the bias of the load for performance monitoring between different nodes is smaller than that in the first embodiment.

<Fifth Embodiment of the Invention> The present invention can be similarly applied to a distributed computer system including a plurality of computers connected by an external network. In this case, a collection process is arranged on each of a plurality of computers belonging to this distributed system, and a collection process is arranged on one of those computers, for example, a computer that manages a network. Of those calculators,
A receiving process is allocated to one of the multiple computers to be monitored, and a part of the multiple computers to be monitored or to any one computer that is different from the multiple computers to be monitored A display process or a storage process may be arranged. It should be noted that it is possible to activate the receiving process on a computer different from the plurality of computers to be monitored. As a result, the communication between each collection process and the collection process or the communication between the collection process and the reception process is performed via the external network, which is different from the first embodiment. However, the performance data can be distributed from the receiving process to a plurality of display processes or storage processes, as in the first embodiment.

<Modifications> The present invention is not limited to the above-described embodiments, but includes various modifications including a plurality of modifications illustrated below. (1) Use of high-speed communication procedure inside parallel computer In the embodiment, it is assumed that communication inside the parallel computer, that is, communication by TCP / IP is used for passing performance data between the collection process 4 and the collection process 3. went. However, in many parallel computers, an internal network that can transfer a plurality of data in parallel via different paths is used. For communication between nodes via such an internal network, there are many cases in which an internal communication procedure that is lighter and faster than TCP / IP is used. In such a computer, the first embodiment
If a method of transferring the performance data collected by other nodes to the collection process activated by one or some of the multiple nodes in the computer, such as 2 and 3, is used for communication between the nodes. It is effective to use a high-speed internal communication protocol peculiar to a parallel computer in order to speed up the transfer of performance data.

In this case, in the collection process 3 shown in FIG. 3, instead of the input analysis routine 205 and the output control routine 206, a set of an input analysis routine and an output control routine corresponding to the internal communication, and an input corresponding to the external communication. A set of analysis routine and output control routine is prepared, and depending on whether internal communication or external communication is performed,
One of these sets is used. I / O buffer 201
Is used by the former and the input / output buffer 203 is used by the latter. For example, the data sent from the collection process 4 to the input / output buffer 201 by the high-speed internal communication procedure is received by the input analysis routine corresponding to the high-speed internal communication for the collection process 4, and is once passed to the internal processing routine 207. Be done. The internal processing routine 207 calls the output control routine 206 corresponding to the external network communication for communicating with the receiving process 15 and passes the data. The output control routine 206 is TCP
The data is transmitted to the receiving process 15 by a protocol such as / IP which enables communication with the monitoring computer 11.
With this modification, all the nodes of the parallel computer are TCP / I
It is possible to reduce the monitoring load on the parallel computer 1, as compared to the case where a relatively heavy communication protocol such as P is used.

(2) Designation of Items to be Collected by Receiving Process In the first embodiment, the performance data for each item to be collected by the collecting process is predetermined, but these items may be specified by the receiving process. To this end, when the system administrator activates the reception process, the computer 11 that activates the reception process may input information designating these data.

(3) Selection of Partial Item-Specific Performance Data in Receiving Process In the first embodiment, a plurality of item-specific performance data collected in the collecting process (for example, CPU utilization rate, memory utilization rate, magnetic field). Disk storage device access count, communication count)
Was always transferred to the display process 16 via the collection process 3 and the reception process 15. On the other hand, if only the items necessary for display are selected by the receiving process 15 and transferred to the display process 16, the amount of transferred data can be reduced. To achieve this, display process 1
6 is activated by the user, and immediately after the connection is established to the receiving process 15 (step 562 (FIG. 6B)), the display process 16 notifies the receiving process 15 of performance data to be displayed. To do. Reception process 15
Stores which display process requires which itemized performance data. When the internal processing routine 308 in the receiving process 15 copies the performance data from the input / output buffer 304 to the input / output buffer 301 connected to any of the display processes in the memory, instead of copying the entire received performance data, Select and copy only the item-specific performance data that is stored as necessary.

(4) Selective collection of performance data by item In the first embodiment, the collection of performance data is requested from the reception process to the collection process independently of the activation of the display process or the accumulation process. However, it is also possible to configure the receiving process to issue this request after any display or storage process has been activated.
At that time, the monitoring process is performed by letting the receiving process notify the receiving process of the item-specific performance data requested by the user and collecting the item-specific performance data notified by those processes started by the receiving process. Can be requested. With this method,
When a new display process or storage process is started, the performance data for each item requested by the receiving process is collected so that the performance data for each item requested by the newly started process is additionally collected. It is desirable to update. By doing so, since only the minimum required item-specific performance data is collected, the amount of collected performance data is reduced.

(5) Batch Transfer of Collected Data at Plural Time Steps In the first embodiment, the data is transmitted from the collecting process 4 to the displaying process 16 at each time step of collecting data. As a modified example, a method is conceivable in which performance data at a plurality of time steps is accumulated in the collection process 4 and these are collectively sent out by one data transmission. The modification will be described below. In FIG. 2, when the output control routine 103 of the collection process 4 receives data from the counter control / readout routine 104, the output control routine 103 stores the data in the last portion of the input / output buffer 101 at that time, and stores the record in the input / output buffer 101. Remember the number. When the number of records in the input / output buffer 101 reaches a predetermined value, the output control routine 103 collects the data stored up to that point and transmits it as one data to the collection process 3. The collecting process 3 and the receiving process 15 operate in the same manner as in the first embodiment described above. Display process 16 that finally received this data
Refer to the internal clock 407 in FIG.
Data for one time is read from the data in the transmission / reception buffer 404 at regular time intervals and displayed. After processing all the data in the transmission / reception buffer 404, the display process 16 transmits a data request to the reception process 15. There is no change in the operation of the accumulation process 17,
The data is stored as in the first embodiment. This makes it possible to reduce the number of times of transfer of the collected data and reduce the communication load.

[0071]

According to the present invention, even if the number of monitoring processes (display process or storage process) using the performance data collected from the monitored computer increases, the load on the monitored computer itself is increased. Is almost never increased.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of a parallel computer performance monitoring system according to the present invention.

FIG. 2 is a module-related diagram of a collection process in a parallel computer performance monitoring system.

FIG. 3 is a module-related diagram of a collection process in the parallel computer performance monitoring system.

FIG. 4 is a module related diagram of a reception process in the parallel computer performance monitoring system.

FIG. 5 is a module related diagram of a display process / storage process in the parallel computer performance monitoring system.

6A is a flowchart showing a part of a processing procedure of a parallel computer performance monitoring method adopted in the system of FIG. (B) is a flowchart which shows another part of the processing procedure of the parallel computer performance monitoring method employ | adopted by the system of FIG.

FIG. 7 is a diagram showing an example of a display screen of performance data.

FIG. 8 is a diagram showing another example of a display screen of performance data.

FIG. 9A is a diagram showing an example of the format of a header record of a block stored in a storage device by a storage process. (B)
FIG. 3 is a diagram showing an example of a format of a normal data record of a block that the storage process stores in a storage device. FIG. 6C is a diagram showing an example of the format of a block end record of a block stored in a storage device by a storage process. FIG. 6D is a diagram showing an example of the format of a file stored in a storage device by a storage process.

FIG. 10 is a flowchart showing a procedure for redisplaying accumulated data.

FIG. 11 is a diagram showing an example of a control screen for controlling the display function of accumulated log data.

FIG. 12 is an overall configuration diagram of a parallel computer to which the performance monitoring method according to the present invention is applied.

FIG. 13 is an overall configuration diagram of another parallel computer to which the performance monitoring method according to the present invention is applied.

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] April 8, 1997

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Fig. 6

[Correction method] Change

[Correction contents]

6 is a parallel computer performance model used in the system of FIG .
The flowchart which shows the process sequence of a nitering method.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Nobutoshi Sagawa 1-280, Higashi Koikekubo, Kokubunji City, Tokyo Inside Central Research Laboratory, Hitachi, Ltd. (72) Inventor Tadashi Ota 5-2-1, Mizumizumoto-cho, Kodaira-shi, Tokyo No. Hitate Super LSI Engineering Co., Ltd. (72) Inventor Susumu Yamaga 5-20-1 Kamimizuhonmachi, Kodaira-shi, Tokyo Inside Hitate Super LSI Engineering Co., Ltd.

Claims (26)

[Claims] 1. A computer network having a plurality of computers including a computer system to be monitored and a computer connection network for connecting them, wherein the computer system to be monitored provides performance data of the computer system at a plurality of timings. The performance data collected by the monitored computer system is monitored by the receiving process activated on one computer of the multiple computers other than the monitored computer system. Received from the target computer system via the computer connection network, among the plurality of computers, equal to or different from the one computer, started on a plurality of computers other than the monitored computer system, For each of the multiple usage processes for using performance data, Computer performance monitoring method comprising the step of transferring the performance data by the receiving process. 2. The plurality of utilization processes have a plurality of display processes for displaying performance data on a display device connected to the plurality of computers on which the utilization processes are activated. Computer performance monitoring method. 3. The performance data includes performance data of a plurality of items, and the method is a part of the performance data of a plurality of items included in the performance data distributed to each display process. The computer performance monitoring method according to claim 2, further comprising the step of displaying the different performance data on a display device connected to a computer on which the display process is activated by the display process. 4. The method further comprises the step of receiving, by the receiving process, a performance data transfer request sent from each display process, and the step of transferring the performance data from the receiving process to each display process is performed by the receiving process. 4. The computer performance monitoring method according to claim 3, which is executed in response to the transfer request sent from the display process. 5. The step of transferring performance data to each display process includes the step of already transmitting a performance data transfer request to the receiving process each time the receiving process receives the performance data from the monitored computer system. If there is one display process that has already sent the performance data transfer request, the received performance data is transferred to the detected display process, and the performance data is transferred. 5. The computer performance monitoring method according to claim 4, further comprising the step of transferring the received performance data to each of the detected plurality of display processes when a plurality of display processes which have already sent the transfer request are detected. . 6. The computer performance monitoring method according to claim 4, wherein some of the item-specific performance data displayed by one of the plurality of display processes is item-specific performance data selected by a user of the display process. . 7. The partial item-specific performance data displayed by one of the plurality of display processes is item-specific performance data predetermined for the display process.
The described computer performance monitoring method. 8. The computer performance monitoring method according to claim 3, wherein some of the item-specific performance data displayed by one of the plurality of display processes is item-specific performance data selected by a user of the display process. . 9. The performance data according to a part of items displayed by one of the plurality of display processes is performance data according to items predetermined for the display process.
The described computer performance monitoring method. 10. The step of transferring performance data to each display process, the transfer request sent from each display process requesting the transfer of some item-specific performance data is received by the receiving process, and the transfer request is received. Each time the receiving process receives the performance data from the monitored computer system after the transmission of the, the above-mentioned request requested by the transfer request among the plurality of item-specific performance data included in the received performance data. 3. The computer performance monitoring method according to claim 2, further comprising the step of selecting a part of the item-specific performance data by the receiving process and transferring the selected part of the item-specific performance data to the display process by the receiving process. 11. The method further comprises the step of sending a performance data transfer request from the receiving process to the monitored computer system, wherein the receiving step includes sending to the monitored computer system after sending the transfer request. The computer performance monitoring method according to claim 2, comprising the step of receiving performance data sent from the computer. 12. The computer performance monitoring method according to claim 11, wherein the step of transmitting the performance data transfer request is executed regardless of whether or not any of the plurality of display processes is activated. 13. The computer performance monitoring method according to claim 11, wherein the step of sending out the performance data transfer request is executed after any one of the plurality of display processes is activated. 14. The method according to claim 2, further comprising the step of notifying the computer system to be monitored of the plurality of item-specific performance data to be collected from the receiving process before any of the plurality of display processes is activated. The described computer performance monitoring method. 15. The method further comprises the step of inputting information for determining a plurality of item-by-item performance data to be collected into the one computer on which the reception process is started by an operator, and the step of notifying includes: 15. The computer performance monitoring method according to claim 14, which is executed depending on the input information. 16. A transfer request for specifying a plurality of item-specific performance data to be transferred, which is sent by the receiving process when one of the plurality of display processes is first activated.
3. The computer performance monitoring method according to claim 2, further comprising a step of requesting the monitoring target computer system to collect the plurality of item-specific performance data received by the reception process and designated by the transfer request from the reception process. 17. The receiving step comprises: when the computer system to be monitored includes a plurality of nodes each including at least one processor, and at least one node is connected to the computer connection network. There is the step of receiving performance data from one node, where the collection process launched on each node collects the performance data of that node and the collection process launched on the one node The computer performance monitoring method according to claim 1, wherein the computer system to be monitored is programmed so as to collect the performance data collected by the collection process activated in each of the plurality of nodes. 18. The computer system to be monitored further has an internal network for connecting the plurality of nodes, which is different from the computer connection network, and the collection process started on each node from the collection process. 18. The computer performance monitoring method according to claim 17, wherein the transfer of the performance data collected by the node from the process to the collection process is performed via the internal network. 19. From the harvesting process of each node,
Transferring the performance data collected by the node to the collection process is performed via the internal network and according to the internal communication procedure defined for the computer system to be monitored, and the internal communication procedure is 19. The computer performance monitoring method according to claim 18, which is different from a communication procedure defined for the computer connection network that connects the plurality of computers to each other. 20. In the receiving step, the plurality of nodes are divided into a plurality of partitions each including a plurality of nodes, and at least one of the plurality of nodes included in each partition is connected to the computer connection network. In this case, the performance data of the plurality of nodes included in the one partition are collectively received from the one node included in one of the plurality of partitions. The collection process started on each of the plurality of nodes belonging to the partition collects the performance data of that node, and the collection process started on the one node of each partition is connected to the plurality of nodes belonging to the partition. To collect the performance data collected by the collection process started for each 18. The computer system to be monitored is programmed.
The described computer performance monitoring method. 21. The other receiving process for receiving the performance data of another one of the plurality of partitions is started in any one of the plurality of computers, and the performance data of the other partition is received. Each of the other plurality of display processes for displaying is started in one of the plurality of computers, and the performance data of the other partition is received by the other receiving process, and the performance of the other partition is displayed. 21. The computer performance monitoring method according to claim 20, further comprising the step of transferring data to each of the plurality of other display processes by the other receiving process. 22. The receiving step is a distributed computer system in which the computer system to be monitored comprises a plurality of computers selected from the plurality of computers connected by the computer connection network, and the plurality of nodes. Each of which includes one of the plurality of selected computers, collectively receives a plurality of performance data for each of the plurality of selected computers from one of the plurality of selected computers. The distributed computer system, the collecting process is activated on each of the plurality of selected computers, and the collecting process is performed on the one computer of the plurality of selected computers. The nature of the computer that the collection process was launched on and the collection process was launched on the selected computers. The computer performance monitoring method according to claim 1, wherein the computer performance monitoring method is programmed to collect performance data. 23. In the receiving step, the computer system to be monitored is a computer system including a plurality of nodes, and when each node is connected to the computer connection network, the receiving step includes: And receiving the performance data of the node from the monitoring target computer system, wherein the collection process started on each node is programmed to collect the performance data of the node. The described computer performance monitoring method. 24. The computer system to be monitored includes a plurality of processors connected to each other, and in the case where at least one of the plurality of processors is connected to the computer connection network, the receiving step comprises: Collecting the performance data of each of the plurality of processors and receiving from the one of the processors, the performance data of the processor by a sampling process activated on each of the plurality of processors. The computer performance according to claim 1, wherein the computer system is programmed to collect the performance data collected by the collection process of each processor by a collection process started on the one processor. Monitoring method. 25. The plurality of utilization processes have a plurality of accumulation processes for storing performance data in a storage device connected to the plurality of computers on which the utilization processes are activated. Computer performance monitoring method. 26. A computer monitoring system for monitoring a computer system to be monitored via a computer connection network, comprising one computer connected to the computer connection network and started on the one computer. And a receiving process, wherein the receiving process receives the collected performance data from the monitored computer system every time the monitored computer system repeatedly collects the performance data of the computer system at different timings. Programmed to transfer the received performance data to each of first and second display processes for displaying performance data, wherein the first display process is equal to or different from the one computer, The second computer is started on the first computer connected to the computer connection network. Shows the process, the first different than the computer, which has been started on the second computer connected to said computer connection network.