JP2004038654A - System for detecting abnormality in parallel computer, and method for detecting the same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain a prompt and adequate response by specifying an abnormal part when abnormality occurs in a parallel computer. <P>SOLUTION: An abnormality detecting system of the parallel computer is used by connecting a plurality of nodes individually including a CPU board to a server via a network. The server comprises an operation state monitor program for monitoring the operation state of each node by fixed period via the network, and is set to output the operation state monitor result signal of each node. Each node comprises a function for normally performing self diagnosis, and is set to normally output the self diagnostic result signal of the operation state in every node by fixed time. The system includes: a switching device for switching the operation state monitor result signal of each node into the self diagnostic result signal of the operation state in each node; and a monitor for displaying the output of the switching device. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an abnormality detection system and an abnormality detection method capable of detecting an abnormal part in a parallel computer using a plurality of CPUs (central processing units) connected to a server via a network.
[0002]
[Prior art]
A conventional abnormality detection system for a parallel computer will be described with reference to FIGS. FIG. 2 is a schematic diagram of a configuration of a conventional parallel computer abnormality detection system. FIG. 3A is an example of a monitor screen for abnormality detection in FIG. 2, and FIG. 2B is an enlarged view of a part of FIG. It is.
[0003]
2. Description of the Related Art In recent years, for science and technology calculations and the like, parallel computers that operate a plurality of small computers in parallel have been used instead of conventionally used large computers such as supercomputers.
[0004]
A parallel computer (Beowolf type) using a plurality of CPUs connected to a server via a network requires a small space when a personal computer for office use, which is sold in the general market, is connected to a network and systemized with a management server. There is a problem of increasing.
[0005]
Therefore, a configuration is adopted in which a plurality of single CPU segments for industrial use, which are usually individually assembled, are connected via the hub 2 and managed by the server 1.
[0006]
FIG. 2 shows an outline of the configuration of such a parallel computer. In FIG. 2, reference numerals 3a to 3n denote a plurality of nodes constituting the parallel computer, and each of the nodes 3a to 3n individually includes a CPU board and a LAN ( It is a CPU segment having a local area network (HDD) signal board, a hard disk drive (HDD), and the like. Each of the nodes 3a to 3n is mounted on a backplane in one box for every several nodes (usually up to about four nodes), and the total number may be ten.
[0007]
Each of the nodes 3a to 3n is connected to the hub 2 by a network 4, and is connected to the server 1 via the hub 2 and is placed under management and control of the server 1.
[0008]
There are software management tools that are open to the public for detecting abnormalities in parallel computers, but those that monitor the individual status of CPUs use each CPU time for their operation, and do The operation of the main program may be hindered.
[0009]
For this reason, an abnormality detection system in a parallel computer as shown in FIG. 2 normally sends a fixed period (for example, several seconds) to each of the nodes 3a to 3n from the server 1 via the hub 2 and the network 4 by the operation status monitoring program of the server 1. The server 1 outputs a polling result signal X, and displays the operation status of each of the nodes 3a to 3n on a monitor screen 5a of a monitor 5 connected to the server 1. This program software runs on, for example, a WWW browser.
[0010]
FIG. 3A shows an example of the monitor screen 5a. The nodes 3a to 3n are polled (operation status monitoring) for each of the boxes A to P to indicate that the nodes 3a to 3n are mounted in the boxes A to P. 3.) A node operation status display 13 showing the result is provided.
[0011]
As shown in FIG. 3B by enlarging and illustrating the four node operation status displays 13 of one box A, the operation status of the CPU board, the LAN signal board, and the HDD for each node is indicated by an LED (light emitting). (light emitting diode). The light emission display of the LED is, for example, blue when the operation rate is less than 50%, yellow when 50 to 80%, red when 80% or more, and black when the system is down. The unit of the operation status display 13 is set finer or larger for one node as a whole according to the system configuration level.
[0012]
[Problems to be solved by the invention]
However, the conventional parallel computer abnormality detection system as described above is based on access from the server 1 side. Therefore, even if a certain node has a system down display (for example, an LED light emission display: black), it is not It cannot be specified whether the problem is caused by the down of each board or HDD in the CPU segment of the node or by the abnormality of the network 4 connecting the node and the hub 2. Therefore, there has been a problem that it is difficult to quickly and appropriately deal with the occurrence of an abnormality.
[0013]
The present invention solves such a problem in the conventional parallel computer abnormality detection system, and can specify an abnormality portion when a parallel computer is abnormal, and can provide a parallel computer abnormality detection system and abnormality detection method capable of promptly and appropriately coping. The task is to provide.
[0014]
[Means for Solving the Problems]
(1) The present invention has been made to solve such a problem, and as a first means, an abnormality detection system in a parallel computer that uses a plurality of nodes each having a CPU board connected to a server via a network. Wherein the server is provided with an operating status monitoring program for monitoring the operating status of the nodes at regular intervals via the network, and is set to output an operating status monitoring result signal of each of the nodes; A self-diagnosis function is set so that a self-diagnosis result signal of the operation status of each individual node is always output at regular intervals, and the operation status monitoring result signal of each node and the operation of each individual node are set. A switching device for switching a self-diagnosis result signal of a situation; and the monitor for displaying an output of the switching device. Provides an abnormality detection system of a parallel computer according to claim.
[0015]
According to the first means, the general operation status of all nodes can be generally known from the result of monitoring the operation status of the server. If a specific node is designated and switched by the switching device, the self-diagnosis of the operation status is performed. Since it is possible to know the specific state of the result, it is possible to monitor the operating state of the entire parallel computer and, when an abnormality is detected once, to grasp the more detailed state and specify the abnormal part easily. .
[0016]
(2) As the second means, the abnormality detection system of the parallel computer of the first means is used. Normally, the switching device outputs an operation status monitoring result signal of each node to the monitor, and When a node indicating an abnormality appears in the operation status monitoring result signal, the switching device outputs the self-diagnosis result signal of the operation status of the node indicating the abnormality to the monitor. I will provide a.
[0017]
According to the second means, the function of the first means is exhibited, and when an abnormality is found in a specific node based on the operation status monitoring result of the server, the switching device checks and compares the self-diagnosis result of the node. This makes it possible to specify whether the abnormal part is in the node itself or in the network.
[0018]
BEST MODE FOR CARRYING OUT THE INVENTION
An abnormality detection system and an abnormality detection method for a parallel computer according to an embodiment of the present invention will be described with reference to FIG. FIG. 1 is a schematic configuration diagram of the abnormality detection system for a parallel computer according to the present embodiment.
[0019]
In FIG. 1, the same parts as those in FIGS. 2 and 3 for explaining the above-described conventional example are denoted by the same reference numerals, and description thereof will be omitted. Differences will be mainly described below.
[0020]
In the abnormality detection system for a parallel computer according to the present embodiment, the nodes 3a to 3n, which are CPU segments including a CPU board, a LAN (local area network) signal board, and an HDD (hard disk drive), are connected to an industrial CPU board. The watchdog timer (Watch Dog Timer) function (a function of performing a self-diagnosis by constantly interrupting) is provided as a standard feature. The watchdog timer function is used to control each of the nodes 3a to 3n (CPU segment). An interrupt program is created so that the result of the self-diagnosis of the operating condition is constantly output to the monitor 5 at regular intervals (for example, every few seconds).
[0021]
In FIG. 1, Ya, Yb, and Yc to Yn are individual self-diagnosis result signals output from the nodes 3a, 3b, 3c to 3n by the watchdog timer function.
[0022]
X is a polling result signal (operation status monitoring result signal) output as a result of polling (operation status monitoring) of each of the nodes 3a, 3b, 3c to 3n by the server 1 described with reference to FIGS. In this embodiment, it is preferable that the protocol for monitoring polling employs HTTP, which has a relatively light load on the network, so that data communication is completed in one shot per monitoring timing.
[0023]
The polling result signal X and the self-diagnosis result signals Ya, Yb, Yc to Yn are input to the switching device 6, which normally selects the polling result signal X and outputs Z to the monitor 5, and the monitor 5 The operating statuses of all the nodes 3a, 3b, 3c to 3n are displayed on the monitor screen 5a of FIG.
[0024]
The switching device 6 designates a specific node 3i and switches the output Z. When the node 3i is designated and receives a switching instruction, the self-diagnosis result signal Yi by the watchdog timer function of the node 3i is output to the output Z. Then, the monitor screen 5a is switched, and the content of the self-diagnosis result signal Yi is displayed on the monitor screen 5a.
[0025]
The display contents of the self-diagnosis result signal Yi are set by the watchdog timer function of each node. However, the display contents do not merely indicate the operation status of the CPU board, the LAN signal board, the HDD, etc. A specific operation state or data can be displayed.
[0026]
The switching device 6 may be an automatic switching device that automatically switches to the self-diagnosis result signal Yi for a node according to a set condition such as when the polling result signal X of any node exceeds a certain range. It may be a switch near the monitor 5 which can be operated by the operator at any time while watching the monitor screen 5a, or a changeover switch attached to the monitor 5, and may be set according to the control level of the abnormality detection system.
[0027]
According to the abnormality detection system for a parallel computer according to the present embodiment as described above, the node operating status displays 13 of all the nodes 3a to 3n based on the polling result of the server 1 are normally displayed on the monitor screen 5a. It is possible to know the general operating status of all the nodes 3a to 3n, and to know the specific state of the self-diagnosis result Yi by specifying the specific node 3i and switching the switching device 6.
[0028]
Then, when an abnormality is found in the node operation status display 13 of the specific node 3i as a result of the polling of the server 1, the switching unit 6 identifies the abnormal point by comparing the self-diagnosis results Yi of the node 3i with the self-diagnosis result. It becomes possible.
[0029]
That is, when the self-diagnosis result Yi of the node 3i is normal when the polling result of the specific node 3i indicates an abnormality, it can be understood that an abnormality has occurred in the network 4.
[0030]
If the polling result of the specific node 3i indicates an abnormality and the self-diagnosis result Yi of the node 3i is also abnormal, it is first known that the node 3i is abnormal.
[0031]
Therefore, while monitoring the operation status of the entire parallel computer by the node operation status display 13 of the polling result of all nodes, once an abnormality is detected, it is easy to grasp the more detailed status and easily identify the location of the abnormality. Thus, it is possible to quickly and appropriately deal with the abnormality.
[0032]
The embodiments of the present invention have been described above. However, the present invention is not limited to the above-described embodiments, and various changes may be made to the specific structure and configuration within the scope of the present invention. It is.
[0033]
【The invention's effect】
(1) According to the first aspect of the present invention, in the abnormality detection system for a parallel computer in which a plurality of nodes each having a CPU board are connected to a server via a network, the server detects the abnormality of the parallel computer. An operation status monitoring program is provided to monitor the operation status of the nodes at regular intervals via a network, and is set to output an operation status monitoring result signal of each node. The node has a function of constantly performing a self-diagnosis. The self-diagnosis result signal of the operation status of each individual node is set to be always output at regular time intervals, and the operation status monitoring result signal of each node and the self-diagnosis result signal of the operation status of each node are provided. And a monitor for displaying the output of the switching device. It is possible to know the general operating status of all nodes based on the server operating status monitoring results, and to know the specific status of the self-diagnosis result of the operating status if a specific node is designated and switched by the switching device. Therefore, once an abnormality is detected while monitoring the operating state of the parallel computer as a whole, it is possible to understand the state of the abnormality in detail and to easily identify the location of the abnormality, and to quickly and appropriately deal with the abnormality. Can do it.
[0034]
(2) According to the second aspect of the present invention, an abnormality detection method for a parallel computer is provided by using the abnormality detection system for a parallel computer according to the first aspect of the present invention. Is output to the monitor, and when a node indicating an abnormality appears in the operation status monitoring result signal of each node, the switching device outputs the self-diagnosis result signal of the operation status of the node indicating the abnormality to the monitor. With such a configuration, the effect of claim 1 is exhibited, and when an abnormality is found in a specific node based on the operation status monitoring result of the server, the switching device looks at the self-diagnosis result of the node and compares the results. It is possible to specify whether the abnormal part is in the node itself or in the network.
[Brief description of the drawings]
FIG. 1 is a schematic configuration diagram of an abnormality detection system for a parallel computer according to an embodiment of the present invention.
FIG. 2 is a configuration schematic diagram of a conventional parallel computer abnormality detection system.
3A is an example of a monitor screen for detecting an abnormality in FIG. 2, and FIG. 3B is an enlarged view of a part of FIG.
[Explanation of symbols]
1 server 2 hub 3a, 3b, 3c to 3n node 3i node 4 network 5 monitor 5a monitor screen 6 switching device 13 node operation status display

Claims (2)

In an abnormality detection system in a parallel computer in which a plurality of nodes each having a CPU board are connected to a server via a network, the server monitors an operation status of each of the nodes at regular intervals via the network. A program is set to output an operation status monitoring result signal of each node, and the node has a function of always performing self-diagnosis, and always outputs a self-diagnosis result signal of the operation status of each of the same nodes at regular time intervals. And a switching device for switching between an operation status monitoring result signal of each node and a self-diagnosis result signal of the operation status of each individual node, and the monitor for displaying an output of the switching device. An abnormality detection system for a parallel computer, comprising: 2. An abnormality detection system for a parallel computer according to claim 1, wherein the switching device normally outputs an operation status monitoring result signal of each node to the monitor, and indicates an abnormality in the operation status monitoring result signal of each node. When the node appears, the switching device outputs the self-diagnosis result signal of the operation status of the node indicating the abnormality to the monitor, wherein the abnormality detection method for the parallel computer is provided.

Images