JPH0934850A - Method for displaying performance data in parallel computer system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To recognize distribution of accurate performance data intuitionally by plotting performance data in the direction of height of an axis orthogonal to a picture simulating each node of parallel operation computers on a plane coordinate. SOLUTION: A picture simulating computers on a plane coordinate is plotted on a display screen 1 and performance data of collected parallel computer system are converted and displayed in the direction of height of an axis orthogonal to the picture. In this case, a minimum unit of display is a node and a network, and a node part is made up of, e.g. a frame 101, a CPU 102, a memory 103, a reception amount 104 and a transmission amount 105, and the network is made up of a 1st communication channel 98, a 2nd communication channel 99 and a router 100 and they are plotted. A thickness in the direction of height of a displayed image in response to the received performance data is changed for each part except the frame 101 to represent the performance data at that time. While expressing a difference from each node of the parallel computer system, a cross reference with an actual indication is easily taken through the display method as above.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for displaying performance data of a parallel computer and a computer connected to the parallel computer via a network.

[0002]

2. Description of the Related Art In order to bring out the performance of a parallel computer,
It is important that the program running on it uses the plurality of processors evenly and does not generate an extremely heavy load (called a bottleneck) on each CPU, I / O, and network. Usually, it is not easy for the user to write such an ideal parallel program from the beginning. Therefore, the procedure is to first create a program that operates logically correctly, observe the operation status on the parallel machine, perform tuning, and finally finish the program that operates at high speed. In order to reduce the effort for that, a tool that allows the user to quickly grasp the operating status of the parallel program is essential.

Conventionally, as a method of displaying performance data of a computer (CPU operating rate, I / O usage status, memory usage status, network usage status, etc.), Hewlett
HP Glanc, a product of Hewlett-Packerd
A display method using a bar graph or a pie chart to represent data at a certain moment and a line graph to represent time series data has been known, as is performed in e Plus. Alternatively, Intel
There is also a method of drawing the diagram corresponding to the parallel computer on the computer screen and changing the color of the corresponding part according to the high and low of the performance data to display the operating state of the parallel computer, as is done in Parade of the company. It was

[0004]

In the above prior art, it was difficult to display the performance data in association with each node of the parallel computer which actually collects the performance data. In addition, it is not possible to easily express the differences in peripheral devices such as the disk capacity of each node and the differences in CPU performance, and it is difficult to easily display several types of performance data collected from each node at the same time. there were.

Even when the time change of the performance data is displayed as a graph focusing on a certain kind of performance data, the visibility is remarkably decreased as the number of displayed nodes increases, and the performance data at a certain time and the performance changing with time When viewing the data at the same time, there was a problem in that it was necessary to display multiple screens for drawing each graph to grasp the information.

It is an object of the present invention to provide a method of displaying the collected performance data of a parallel computer so that it can be grasped as easily as possible.

[0007]

In order to achieve the above-mentioned object, the method of the present invention draws a picture simulating each node of a parallel computer in a plane coordinate form, and performs performance in the height direction of an axis orthogonal to it. Display as a three-dimensional graph with data assigned.
Further, when the time change of a certain type of performance data is displayed as a graph, it is displayed as a three-dimensional graph in which nodes, time, and performance data are respectively assigned to the three orthogonal axes.

[0008]

According to the above means, the performance data is drawn in the height direction of the axis orthogonal to the picture simulating each node of the parallel computer in the plane coordinates, so that the correspondence between each node and the performance data becomes easy. , It is possible to intuitively recognize the accurate distribution of performance data. Also, when focusing on a certain type of performance data, by integrating and displaying the graphs that were conventionally displayed individually, it is possible to easily understand each node, time, and performance information on a single screen. It becomes possible.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described with reference to the drawings. The present invention uses performance data (C
The present invention relates to a method of displaying the PU operation rate, I / O usage status, memory usage status, network usage status, etc.). Specific examples of the display are shown in FIGS. 1 and 2.

First, FIG. 1 shows that the performance information of the sampled parallel computer is drawn on the display screen 1 as a plane-coordinate image of the computer, and the performance data is plotted in the height direction of the axis orthogonal to it. Is an example in which is converted and displayed.

FIG. 3 shows the details of a part thereof. The minimum unit of display is a node and a network. In this example, the node portion is the frame 101, the CPU 102, the memory 10.
3, a reception amount 104 and a transmission amount 105, and the network includes a first communication path 98, a second communication path 99, and a router 1.
The drawing is composed of 00. Frame 1
Each part except 01 changes the thickness in the height direction of the display image according to the input performance data, and shows the performance data at that time. The shape and configuration can be changed by inputting from a file described later. With this display, it becomes possible to easily correspond to the actual display while expressing the difference of each node of the parallel computer.

Next, as shown in FIG. 2, a node, a time, and a performance are assigned to three orthogonal axes, and designated items are selected from the collected performance data of the parallel computer, and these are integrated and displayed. It is an example. 106 to 112 are examples in which items (for example, CPU utilization rate) designated from the performance data obtained from each node of the parallel computer are displayed. Each item displayed below will be described.

Reference numeral 106 shows the maximum value among the performance data currently displayed, and 107 shows the number of the node that is displaying. 108 and 111 indicate the times at which the data were collected. Reference numeral 109 draws performance data in the height direction of an axis orthogonal to the plane formed by the node and the time axis. Similarly, 110 also assigns past performance data to the height and draws it in a strip shape so as to connect the points. Reference numeral 112 denotes an auxiliary line that is similarly drawn so as to allocate the maximum value of the performance data currently displayed in the height direction and to easily recognize the state of the entire performance data.

FIG. 4 shows a part of the details. The display consists of a mainframe (10) showing the performance data at the current time and a train (11) showing the past performance data. With this display method, it is possible to instantly recognize each node, time, and performance data. It is also possible to switch to a conventional two-dimensional graph display method.

When this display is aligned with the time axis in the normal direction of the screen, the performance is displayed by a bar graph which is a conventional display method as shown in FIG. 5. Similarly, when the axes of the nodes are aligned, As shown in 6, the line graph is a conventional display method. Further, when the axes of the performance data are aligned, the change in the value can be displayed by color as shown in FIG.

FIGS. 8 and 9 show examples of display window configurations including the above-described display method. The data collected by the parallel computer is processed according to the user's designation and displayed as 120 or 123. In the areas indicated by 121 and 124 in the window, special functions such as history display and enlarged display of performance data are designated, and 122 and 1
In the part indicated by 25, the operation of the currently displayed one such as changing the viewpoint and starting / stopping the display is designated.

FIG. 10 shows the configuration of the entire system. The following is a description of this figure. A computer 13 that activates the data collection unit 132 on a plurality of nodes 131 of the parallel computer indicated by 130 and displays the data via the network 133.
4 to the input device 135. On the computer 134 side, the data collection unit 136 collects the sent performance data and relays the performance data to the display unit 138 which is currently displaying via the network. The relayed performance data is displayed on the display 140 as shown in FIGS. 8 and 9 above, using the information in the system configuration file 139. In addition, the performance data is saved in log file 1 according to the specification.
37.

FIG. 11 and subsequent figures show the details of the processing algorithm for displaying the graph on the windows shown in FIGS. FIG. 11 shows the processing of the data collection unit 132. Each node of the parallel computer performs the following processing (150) until a termination signal is sent at regular time intervals. First, the buffer is initialized (151), the CPU performance data is collected (152), the memory performance data is collected (153), and so on.
4), performance data of receiving system (155), disk I / O
Performance data (156), performance data of the number of execution instructions (1
57), cache performance data (158) and network performance data (159) are collected and data is transmitted (160).

Each performance data can be obtained by issuing a system call to the operating system running on each node. When the operating system is UNIX, pstat, for example, may be used as the system call.

FIG. 12 shows the processing of the data collection unit 136 which collectively processes the performance data sent from the data collection unit. The data collection unit first performs a connection process (170) with the data collection unit that has been started in advance at the time of startup, and then performs initial settings such as a data collection interval and collection items (171). After that, the following processing is performed until the termination signal is sent.

First, the data transmission source is determined (17)
3). If the data source is the data collection unit,
Whether or not there is a transmission request for the currently opened window is checked (174), and data for the requested one is transferred (175). If it is a data display unit, the type of request is first determined (176), if it is a connection request, connection processing (177), and if it is a disconnection request, disconnection processing (17).
8) In the case of a performance data request, the list referred to when transmitting the performance data is updated (179), and in the case of a parameter change request, the parameter to be changed is transmitted to the data collection unit.

The performance data is displayed on the display unit 13 by the above processing.
8 and is displayed on the display by the display method of the present invention. Regarding the processing algorithm on the display unit 138,
First, the display method of FIG. 1 will be described with reference to FIGS. 13 to 19 and FIG. 24, and the display method of FIG. 2 will be described with reference to FIGS.

First, the outline of the processing for the display of FIG. 1 will be described. As shown in FIG. 13, when the window for displaying in FIG. 1 is activated, first, the system configuration is read from the file and data for reproducing the configuration of each node is created (200). FIG. 14 shows an example of the contents of the file to be read. The file is composed of a node identifier, a reference coordinate, a format, and shape data forming a node. Next, an object for storing the information of each node is created (201), and the data read from the file is stored (202).

FIGS. 15 and 24 show the data structure of the object that stores the information. For nodes, the coordinates, width, depth, etc. of the reference point of the shape forming the node
It consists of node identifiers and performance data, and the network has the same structure.

The window display is continued until the quit button is pressed (203). When the start button (122 in FIG. 8) is pressed (204) and stopped (stop)
The following processing is repeated until the button (122 in FIG. 8) is pressed (205). First, the parameters for display are read (206), and performance data is requested to the data collection unit (207). The performance data is read (208), the display data is created (210) and the drawing (214) is repeated for the node from the viewpoint direction to the depth direction (209).
Similarly, for the network in the depth direction, (212) display data creation (213) and drawing (214) are repeated. If there is a request to display numerical data (215)
Also displays this in the window (216). The above is the outline of the display.

FIG. 16 shows details of the above-mentioned portion 214. First, coordinate conversion and drawing of the node frame 101 are performed (250). Next, the performance data of the memory 103 is read (251), and coordinate conversion and drawing are performed (252). Next, the maximum value of the memory is assigned in the height direction, data of auxiliary lines to be drawn is created to easily recognize the state of the performance data (253), and coordinate conversion and drawing are performed (254). C
The PU / reception system / transmission system also performs the same processing (255-2
62).

Next, details of the coordinate conversion and drawing from the above-mentioned processing will be described with reference to FIGS. 17 to 19. FIG.
18 and 18 show processing of drawing a rectangular parallelepiped and a triangular prism which are the minimum drawing units in this example, and FIG. 19 shows coordinate conversion.

FIG. 17 shows a process of drawing a rectangular parallelepiped. The processing will be described in order. First, the color data is read from the file (300). Next, the shape data of the object created by the process 201 shown in FIG.
Eight three-dimensional coordinates of a rectangular parallelepiped are created from the display data created in 3 (301). Using the set angle of viewpoint, the point coordinates obtained in 301 are converted into a two-dimensional plane projected (302). This conversion is shown in FIG.
More on this later. Next, a list is created in order from the surface at the back when each surface of the rectangular parallelepiped is viewed from the line-of-sight direction (303).
The drawing color is calculated from the inner product of the surface and the line of sight and the color data in the order of the created list (304), and finally the surface is drawn (30).
5).

FIG. 18 shows a process of drawing a triangular prism. The processing is as follows. Similar to the processing in FIG. 17, first, color data is read (310). Then 201
The six point coordinates of the triangular prism are created from the shape data of the object created by the process and the display data created by 210 or 213 (311). The point coordinates obtained in 311 are converted into a projection onto a two-dimensional plane using the set viewpoint angle (312). This conversion processing also uses the one shown in FIG. Next, a list is created in order from the surface at the back when seen from the line-of-sight direction (313). The drawing color is calculated from the inner product of the surface and the line of sight and the color data in the order of the created list (3
14) Finally, the surface is drawn (315).

Referring to FIG. 19, the coordinate conversion process performed in the above steps 302 and 312 will be described in detail. In this conversion, the viewpoint is defined as an angle from each of the Y and Z axes, and the three-dimensional coordinate data is mapped to a two-dimensional plane by using them.
The flow will be described below. The three-dimensional point coordinates to be converted are read (330). First, rotation coordinate conversion around the Z axis is performed using the angle of the viewpoint with respect to the Z axis (3
31). Next, rotation coordinate conversion about the Y axis is performed using the angle of the viewpoint with respect to the Y axis (332). The calculated y and z components of the three-dimensional coordinate are respectively the new x and y components of the two-dimensional coordinate.
The conversion process ends by substituting the component (333).

Details of the display method shown in FIG. 2 will be described with reference to FIGS.

First, the outline of the processing for the display of FIG. 2 will be described. As shown in FIG. 20, when the window for displaying in FIG. 2 is activated, first, the system configuration is read from the file and data for reproducing the configuration of each node is created (350). The structure of the file to be read is shown in FIG. 14 by a specific example. File is a node identifier
It consists of standard coordinates and formats and shape data that forms nodes. Next, an object for storing the information of each node is created (351), and the data read from the file is stored (352).

FIG. 21 shows the data structure of the object that stores the information. Quit window
The process is repeated until the button is pressed (353). First, the item currently displayed is read (354). Until the start button is pressed (355) and the stop button is pressed (3
56), the following processing is repeated. First, the parameters for display are read (357), and performance data is requested to the data collection unit (358). Read performance data (35
9) and nodes from the viewpoint direction to the depth direction (36
0) Display data creation (361) and drawing (362) are repeated. If there is a request to display numerical data (36
3) also displays this in the window (364). The above is the outline of the display.

FIG. 22 shows details of the 362 portion. First, the color that serves as the display reference is read (37
0), the current performance data is read and a value is set in the buffer (371). Next, as long as there is data in the buffer (372), the following processing is performed. First, the direction of the line of sight is determined (373). If the line of sight is the front, performance data is drawn by a bar graph (374). In the case of the side surface, the performance data is drawn by a line graph (375). In the case of the upper surface, the performance data is drawn by changing the color gradation (37
6). In other cases, if it is the performance data at the current time, coordinate conversion and drawing of 10 are performed (377). If it is performance data at a past time, coordinate conversion and drawing of 11 parts are performed (378). Finally, 112 is drawn to make it easy to recognize the maximum value of the performance data currently displayed. The drawing of 10 is performed using the processing described in FIG.

Drawing of the portion 11 is performed as shown in FIG. First, read the standard color data (40
0). The data of the coordinates, width, and depth of the reference point of the surface to be drawn is read (401). The following processing is performed in order from one unit time step before the oldest data until there is no performance data in the buffer (402). First, it is checked what time step the drawing surface is from the current time, and the offset value of the brightness from the reference color is calculated (40
3).

The reference point of the surface to be drawn is calculated from the information obtained in 401 (404). Further, the coordinate values of the four points forming the surface are calculated from the performance data and the information obtained in 401 (40
5). The drawing color of the surface is calculated from the inner product of the offset value and the line of sight calculated in 403 (406). As described above, the surface is drawn using the obtained coordinate values and color of the point (407).

By the two kinds of display methods described above, for example, by drawing the performance data on the plane in the height direction of the axis orthogonal to the picture simulating each node of the parallel computer, Can be recognized intuitively, and when one type of performance data is focused on and displayed, by integrating and displaying the graphs that were conventionally displayed individually, It becomes possible to easily grasp each information of node, time, and performance on the screen.

[0038]

According to the present invention, since the performance data is drawn on the plane in the height direction of the axis orthogonal to the picture simulating each node of the parallel computer, the correspondence between each node and the performance data can be facilitated. It is possible to intuitively recognize the distribution of performance data on parallel machines. In addition, when focusing on a certain type of performance data and displaying it, the graphs that were individually displayed in the past are integrated and displayed so that the node
It is possible to easily grasp each time / performance information.

[Brief description of drawings]

FIG. 1 is a diagram showing a drawing state of performance data according to an embodiment of the present invention.

FIG. 2 is a diagram showing a drawing state of performance data according to an embodiment of the present invention.

3 is a detailed explanatory view of a part of FIG. 1. FIG.

FIG. 4 is a detailed explanatory view of a part of FIG.

5 is a diagram showing a display example when the viewpoint of the display method of FIG. 2 is in the front.

6 is a diagram showing a display example when the viewpoint of the display method of FIG. 2 is a side face.

7 is a diagram showing a display example when the viewpoint of the display method of FIG.

FIG. 8 is a diagram showing a display example in which the display method of FIG. 1 is displayed in a window.

FIG. 9 is a diagram showing a display example in which the display method of FIG. 2 is displayed in a window.

FIG. 10 is a block diagram showing a configuration example of the entire system for carrying out the method of the present invention.

FIG. 11 is a processing flow chart of a portion for collecting data on a parallel computer.

FIG. 12 is a process flow chart of a part that collects sent data.

FIG. 13 is a process flow diagram showing an outline of a part that performs the display process shown in FIG. 1.

FIG. 14 is an explanatory diagram showing a file that stores configuration data of a parallel computer.

FIG. 15 is an explanatory diagram showing a data structure used when performing the display shown in FIG. 1.

FIG. 16 is a processing flow chart showing details of a portion for performing the display processing shown in FIG. 1.

FIG. 17 is a processing flow chart of coordinate conversion and drawing.

FIG. 18 is a processing flow chart of coordinate conversion and drawing.

FIG. 19 is a detailed processing flowchart of the coordinate conversion part.

FIG. 20 is a process flow diagram showing an outline of a part that performs the display process shown in FIG. 2.

21 is an explanatory diagram showing a data structure used when performing the display shown in FIG. 2. FIG.

FIG. 22 is a processing flow chart showing details of a portion for performing the display processing shown in FIG.

FIG. 23 is a processing flowchart showing processing of coordinate conversion and drawing part.

FIG. 24 is an explanatory diagram showing a data structure used when performing the display shown in FIG. 1.

[Explanation of symbols]

98 ... Display data indicating the first communication path of the network,
99 ... Display data indicating the second communication path of the network,
101 ... Display data indicating frame of node, 102 ...
Display data indicating the CPU portion, 103 ... Display data indicating the memory portion, 104 ... Display data indicating the reception amount, 10
5 ... Display data indicating the transmission amount.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Nobutoshi Sagawa 1-280 Higashi Koigokubo, Kokubunji City, Tokyo Metropolitan Research Center, Hitachi, Ltd. (72) Inventor Tadashi Ota 5-2-1 Kamisuimotocho, Kodaira-shi, Tokyo No. Nititsu Cho LLS Engineering Co., Ltd. (72) Inventor Shunji Takubo 1-280 Higashi Koikekubo, Kokubunji, Tokyo Metropolitan Research Center, Hitachi, Ltd.

Claims (11)

[Claims] 1. The performance information obtained from each constituent processor of the parallel computer is drawn on a plane coordinate in a picture simulating the computer, and the performance information is converted in the direction of an axis orthogonal to the plane coordinate. A method of displaying performance data, characterized by displaying as a three-dimensional graph. 2. The performance data obtained by each processor of the parallel computer is displayed as a three-dimensional graph by allocating nodes, time and performance to three orthogonal axes, respectively. Display method. 3. The display method according to claim 1, wherein the direction of the viewpoint is designated by an angle, and the result of coordinate conversion is displayed according to the designated angle. 4. The performance data display method according to claim 1, wherein the displayed picture is enlarged or reduced by designation. 5. The method for displaying performance data according to claim 1, wherein the position of the displayed graph on the screen can be moved in any direction by designation. 6. The history of the acquired performance data is stored in a log file while displaying the performance data in real time, and the mode is switched from the real time display according to designation, and the past performance data stored in the log file is stored. 3. The method for displaying performance data according to claim 1, wherein the performance data is retrieved and reproduced at arbitrary time intervals at arbitrary time intervals according to the passage of time or in reverse of the passage of time. 7. The method for displaying performance data according to claim 1, wherein the performance data specified in a separate window is displayed numerically while the performance data is displayed graphically in real time. 8. A box line in which the maximum value of the performance data or a theoretical maximum value at a certain time is taken in the height direction, its numerical value, a node identifier, etc. are drawn on the performance data display. The method of displaying performance data according to claim 1 or 2. 9. The performance data according to claim 1, wherein when the display data reaches a preset value, the portion is enlarged or highlighted and a warning is given. Display method. 10. The difference between the respective nodes forming the parallel computer is expressed by changing the display figure and the color according to the respective nodes forming the parallel computer of the picture simulating the computer. The method of displaying the performance data described in 1. 11. When the viewpoint is in the front, performance data at a certain time is displayed as a bar graph, when it is on the side, time-series performance data is expressed as a line graph, and when it is at the top, time series is displayed. The performance data display method according to claim 2, wherein a color is assigned to the performance data to be represented.