JP2001209623A - Management device for parallel computer system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To collectively conduct the operation and management of plural nodes constituting a parallel computer system by a management terminal equipment. SOLUTION: Each of plural nodes is provided with a main processor to be driven by a main power supply of the node to conduct parallel processing, a sub-processor to be driven by an auxiliary power supply of the node to conduct a system control command for managing the main processor and a system control mechanism to be driven by the auxiliary power supply to conduct communication with a management terminal equipment. The management terminal equipment is provided with a system control mechanism for communication with plural system control mechanisms built in plural nodes and a system control interface consisting of connecting plural system control mechanisms of plural nodes to the system control mechanism of the management terminal equipment and capable of transmitting a system control command for managing each main processor from the terminal equipment to the sub-processor.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a management system for a parallel computer system and, more particularly, to maintenance and management of a plurality of nodes constituting a parallel computer system even when main processors of the nodes are not operating. The present invention relates to a technology effective when applied to a management device of a parallel computer system for performing management.

[0002]

2. Description of the Related Art Heretofore, there have been proposed some operation management methods for computer systems composed of a plurality of computers and apparatuses for implementing the methods.

[0003] A centralized operation of a plurality of UNIX machines and a monitor console display for preventing an increase in the load of operation and monitoring that occur when the consoles of a plurality of UNIX machines are integrated into one console are disclosed in Japanese Unexamined Patent Publication No.
No. 2,147,633.

The outline is that a destination table storing destinations of commands for each operation purpose is created in a center console which is a server for centrally operating and managing a plurality of UNIX machines, and commands are executed according to the destination table. Things.

[0005] In a multi-computer system composed of a plurality of computers, a console switching control method in a multi-computer system for preventing erroneous operation when maintenance and operation are performed by switching computers connected by a single system console is described below. It is described in JP-A-5-120247.

[0006] The outline is that a service processor in a plurality of computers is connected to a switching device, a system console is connected to the switching device, and a message output target is output from the system console using an identifier for identifying the computer. A system console is shared by a plurality of computers by sequentially switching computers, and when maintenance and operation are performed by the system console, an identifier of the computer to be operated and a system console Is compared with the identifier of the computer connected to the computer, and when the identifiers match, the operation is executed.

[0007] A message centralized management system for centrally managing messages from the computers constituting the distributed processing system is described in Japanese Patent Application Laid-Open No. Hei 5-20281.

The outline is a system in which a centralized management node is determined in a plurality of computers connected via a network, and the centralized management node centrally manages operation status messages issued by monitored nodes.

[0009]

The present inventor has found the following problems as a result of studying the above prior art.

That is, in the conventional management system for a computer system comprising a plurality of computers, the functions of the network software running on the computers to be managed are used, so that the computers to be managed are not operating. There is a problem that operation management cannot be performed when the operating system is not operating or when the network software is not operating.

In the above-described conventional method of centrally operating a plurality of UNIX machines and using a monitoring console display, the computer to be managed is premised on that UNIX of the operating system is operating. If not, there is a problem that central operation and monitoring cannot be performed from the console display.

In the conventional console switching control method in the multifunction computer system, since a switching device exists between the system console and each service processor, there is a problem that special hardware as a switching device is required. there were.

In the conventional message centralized management system,
Since a message is sent from a plurality of computers to the central management node, when the central management node goes down, the problem that centralized management of messages cannot be performed and the message is transmitted via a network to which the node is connected When the operating system and the network of each node are not activated, there is a problem that the state of each node cannot be managed from the central management node.

An object of the present invention is to manage the operation and management of a plurality of nodes constituting a parallel computer system irrespective of the operation of a main processor executing parallel processing and the operation of an operating system and network software of the main processor. It is an object of the present invention to provide a technique that can be performed by a single device.

Another object of the present invention is to provide a technique that enables a management terminal device to turn on or off a plurality of nodes constituting a parallel computer system at once or individually.

Another object of the present invention is to provide a technique capable of suppressing a rush current of power supply equipment for supplying power to a parallel computer system.

Another object of the present invention is to provide a technique which enables a management terminal to monitor whether a plurality of nodes constituting a parallel computer system are operating normally.

Another object of the present invention is to collectively store the node messages in the management terminal device even when the main processors of a plurality of nodes constituting the parallel computer system stop the operation after outputting the node messages. An object of the present invention is to provide a technology that can be managed.

Another object of the present invention is to provide a technique capable of collectively managing the contents of a main memory and a register when a failure occurs in a plurality of nodes constituting a parallel computer system by a management terminal device. It is in.

Another object of the present invention is to provide a technique capable of collectively resetting main processors of a plurality of nodes constituting a parallel computer system from a management terminal device.

Another object of the present invention is to change a bootstrap device in accordance with an instruction from a management terminal when a failure occurs in a specific bootstrap device of a plurality of nodes constituting a parallel computer system. It is an object of the present invention to provide a technique capable of performing system startup processing of a main processor of a plurality of nodes.

Another object of the present invention is to provide a technique capable of preventing contention between operations of the plurality of management terminal devices when the parallel computer system is managed by the plurality of management terminal devices. .

Another object of the present invention is to provide a technique capable of remotely managing the operation of a parallel computer system.

The above and other objects and novel features of the present invention will become apparent from the description of the present specification and the accompanying drawings.

[0025]

SUMMARY OF THE INVENTION Among the inventions disclosed in the present application, the outline of a representative one will be briefly described.
It is as follows.

(1) In a management apparatus for a parallel computer system having a management terminal device for managing a parallel computer system in which nodes as a plurality of computers are connected, the plurality of nodes operate in parallel with a main power supply of each node. A main processor that executes processing, a sub-processor that operates on the auxiliary power supply of each node and executes a system control command for managing the main processor, and a system control mechanism that operates on the auxiliary power supply and communicates with the management terminal device The management terminal device includes a system control mechanism that communicates with a plurality of system control mechanisms of the plurality of nodes, a plurality of system control mechanisms of the plurality of nodes, and a system control mechanism of the management terminal device. To control a plurality of main processors of the plurality of nodes. Those comprising a system control interface for transmitting command from the management terminal device to a plurality of sub-processors of said plurality of nodes.

In the management device for a parallel computer system of (1), the system control command issued from the management terminal device is transmitted via the management terminal device constituting the system control interface and the system control mechanism of the plurality of nodes. A system control command sent to the sub-processors of the plurality of nodes and sent to the sub-processor,
The sub-processor executes the operation management of the main processor.

In a conventional management apparatus for a parallel computer system, a general-purpose operating system running on a main processor that executes parallel processing, which is a normal task of the parallel computer system, or a network operating under the management of the operating system We use software to manage the operation of multiple nodes that make up the parallel computer system.

For this reason, when using the conventional parallel computer system management device, the main processors of a plurality of nodes constituting the parallel computer system to be managed operate normally, and the general-purpose operating system is used. It is a precondition that network software is running or when the main processors of the plurality of managed nodes are not operating or the general-purpose operating system or network software is not operating, for example, in parallel. The state before the start of operation when the power of the computer system is not turned on, or the state where the operating system or network configuration is changed and it is trying to confirm whether it operates normally, or
For example, when investigating the status of a specific node that has stopped operating due to the occurrence of a failure, the management device of the conventional parallel computer system cannot be used.

Therefore, in the management apparatus for the parallel computer system (1), a sub-processor and a system control mechanism operable irrespective of the operation state of the main processor include:
The system control mechanism of each node is connected to the system control mechanism of the management terminal device via a device such as a network concentrator, provided for all nodes constituting the parallel computer system of (1).

The sub-processor and the system control mechanism provided in each of the plurality of nodes operate on an auxiliary power supply different from the main power supply used by the main processor,
The system control mechanism communicates with the management terminal using network software operated by the main processor and network cables and communication cables different from communication cables used by the network software.

Therefore, even if the main processor is not operating, it is possible to control the main processor from the management terminal device if the sub-processor and the system control mechanism are operating by the auxiliary power supply.

As described above, according to the management apparatus for the parallel computer system, the network device operates on the auxiliary power supply of each node and uses network software and communication cables different from the network software and communication cables used by the main processor. Transmitting a system control command from the management terminal device to a system control mechanism that communicates with the management terminal device, and executing the system control command on a sub-processor that operates on the auxiliary power supply. Irrespective of the operation of the main processor executing the parallel processing and the operation of the operating system and network software of the main processor, the operation management of a plurality of nodes constituting the parallel computer system is collectively performed by the management terminal device. To do Possible it is.

(2) In the management apparatus for a parallel computer system of (1), the sub-processors of the plurality of nodes have a function of turning on or off a main power supply of the nodes, and the management terminal device includes Means for transmitting a system control command for turning on or off the main power supply collectively or individually to the sub-processors of the node.

In the management device for the parallel computer system of the above (2), the main processor operates on the main power supply, so that the main processor is turned on or off by the sub-processor, thereby turning off the power supply to the main processor. The dosing can be controlled.

Further, the management terminal device transmits the system control command for turning on or off the main power as a packet specifying a destination to all nodes or a specific node via the system control interface. To all nodes at once or individually.

In the management device for a parallel computer system of (2), since the sub-processors and the system control mechanism of the plurality of nodes are operated by the auxiliary power supply, even if the main power supply is not turned on to the main processor. , Operable.

As described above, according to the management apparatus for the parallel computer system, the main power of the plurality of nodes is turned on or off in accordance with an instruction from the management terminal device. The power can be turned on or off collectively or individually by the management terminal device.

(3) In the management device for a parallel computer system according to (2), the management terminal device transmits a system control command for individually turning on the main power of the plurality of nodes at predetermined time intervals. Means for individually transmitting to the sub-processors of the plurality of nodes.

When the main power supply of a plurality of nodes constituting the parallel computer system is turned on at the same time, an excessive rush current flows in the power supply equipment for supplying power to the main power supply, and a load is imposed on the power supply equipment. In the management device for the parallel computer system of (3), the power-on time of the plurality of nodes is shifted for each node so that the inrush power of the power supply equipment is suppressed.

This is achieved by transmitting a system control command for turning on the main power from the management terminal device at a preset time interval as a packet specifying a transmission destination to a specific node via the system control interface. It is performed by

As described above, according to the management apparatus for the parallel computer system, the instruction to turn on the main power to the plurality of nodes is issued at a predetermined time interval, so that the power is supplied to the parallel computer system. The inrush current of the power supply equipment to be supplied can be suppressed low.

(4) In the management device for a parallel computer system according to (1), the management terminal device transmits a specific system control command to a sub-processor of a specific node of the plurality of nodes, and sets a predetermined system control command. When a normal response to the specific system control command is not received within a time period, a means is provided for determining that an abnormality has occurred in the specific node.

In the management device for a parallel computer system of the above (4), the specific system control command is transmitted from the management terminal device to the sub-processor via the system control interface, and the transmitted specific specific command is transmitted. When a system control command is executed by the sub-processor, an execution result of the specific system control command may not be obtained due to an abnormality of the main processor.

In such a case, the management terminal of the management device for the parallel computer system of (4) waits for a response to the specific system control command for a preset time, and If a response indicating that the specific system control command has been normally executed is not received within the set time, it is determined that an abnormality has occurred in the specific node.

As described above, according to the management apparatus for the parallel computer system, it is checked whether or not a normal response to a specific system control command from the management terminal apparatus is received within a predetermined time. It is possible to monitor with the management terminal whether the plurality of nodes constituting are operating normally.

(5) In the management apparatus for a parallel computer system of (1), the system control mechanism of the plurality of nodes accumulates node messages, which are messages output by a main processor or a sub-processor of the node during operation. Means for reading node messages stored in the system control mechanism of the node.

The main processors of a plurality of nodes constituting the parallel computer system output various node messages at each processing stage.

For example, when a main processor of a plurality of nodes constituting the parallel computer system finds an inconsistency in a file system during system startup processing, it outputs a specific node message and starts repairing the file system. I do. If the main processor fails to restore the file system, it outputs a node message indicating that the restoration of the file system failed, and the system startup process ends abnormally.

When the main processors of a plurality of nodes constituting the parallel computer system detect an unrecoverable failure during the operation after the system startup processing has been completed normally, the contents of the failure called a panic message and the occurrence of the failure occur. Outputs a node message containing the contents of the location, etc.,
The main processor that has detected the unrecoverable fault usually stops the operation due to a system down immediately after outputting the panic message to a specific display device.

In such a case, it is necessary to examine the contents of the node message and remove the cause of the abnormal termination of the system start-up process or the system down, but the main processor has already stopped its operation. Therefore, the management device of the conventional parallel computer system cannot transmit and display the node message to the management terminal device.

Therefore, in the management device for the parallel computer system of (5), the node message output by the main processor or the sub-processor of the node at the time of operation is stored in the system control mechanism of the plurality of nodes, and the management terminal device By reading the node message stored in the system control mechanism of the node, the node message is displayed on the management terminal device, and the content of the node message can be examined by the management terminal device. I have.

As described above, according to the management apparatus for the parallel computer system, the node message output by the main processor or the sub-processor of the specific node during operation is stored, and the management terminal device stores the stored node message. Since the message is read, even if the main processors of a plurality of nodes constituting the parallel computer system stop operating after outputting the node message,
The node messages can be collectively managed by the management terminal device.

(6) In the management device for a parallel computer system of (1), the sub-processors of the plurality of nodes include means for referring to and updating the contents of a main memory or a register of the node, and the management terminal device Comprises means for transmitting, to the sub-processors of the plurality of nodes, a system control command for referring to or updating the contents of the main memory or the registers of the nodes.

When a failure occurs in the main processors of a plurality of nodes constituting the parallel computer system, the cause of the failure is examined by referring to the contents of the main memory and registers connected to the main processor, and possible. In such a case, it may be desired to change the contents of the main memory or the register and continue the process interrupted by the failure.

In the above case, the main processor may not be able to operate normally depending on the fault that has occurred. Therefore, the system control command from the management terminal unit issues a sub-processor of the node having the faulty main processor. Refers to or updates the contents of the main memory or register.

As described above, according to the management device of the parallel computer system, the contents of the main memory or the register of the node are referred to or updated by the instruction from the management terminal device. It is possible to collectively manage the contents of the main memory and the register at the time of occurrence of the failure by the management terminal device.

(7) In the management apparatus for a parallel computer system according to (1), the sub-processors of the plurality of nodes include means for resetting a main processor of the node, and the management terminal device includes a sub-processor of the node. The processor includes means for transmitting a system control command for resetting the main processor of the node to the processor.

When it is necessary to reset the main processor in a plurality of nodes constituting the parallel computer system, for example, after upgrading the operating system or other software, or after performing the work for removing the cause of the failure. There is.

In the above case, in the management device for the parallel computer system according to (7), the sub-processors of the plurality of nodes reset the main processors of the nodes according to a system control command from the management terminal device. I do.

As described above, according to the management apparatus of the parallel computer system, the main processors of the plurality of nodes are reset according to an instruction from the management terminal apparatus.
It is possible to collectively reset the main processors of a plurality of nodes constituting the parallel computer system from the management terminal device.

(8) In the management apparatus for a parallel computer system of (1), the sub-processors of the plurality of nodes refer to and update the contents of the main memory of the node, and reset the main processor of the node. And a system control command for referring to and updating a bootstrap device name in a main memory stored in a main processor of the node, and a system control command for resetting a main processor of the node. And means for transmitting a command.

When performing a system startup process of a certain main processor in a plurality of nodes constituting the parallel computer system, changing an operating system or other software to be loaded into a main memory during the system startup process Or, if a failure occurs in the bootstrap device storing the operating system or other software, the bootstrap device used in the system start-up process may need to be changed.

In such a case, in the management apparatus of the parallel computer system of (8), the management terminal refers to the bootstrap device name in the main memory stored in the main processor of the node. After sending a control command to the sub-processor and checking the bootstrap device name in the main memory,
The management terminal device sends a system control command for updating the bootstrap device name in the main memory stored in the main processor of the node to another bootstrap device name to the sub-processor.

The sub-processor, having received a system control command for updating the bootstrap device name in the main memory to another bootstrap device name from the management terminal device, updates the bootstrap device name in the main memory of the node. I do.

Next, the management terminal device sends a system control command for resetting the main processor of the node to the sub processor, resets the main processor of the node, and updates the main processor by another updated bootstrap device. Performs system startup processing of the processor.

As described above, according to the management device of the parallel computer system, the bootstrap path information in the main memory of the plurality of nodes is changed according to the instruction from the management terminal device, and the main processor is reset. ,
When a failure occurs in a specific bootstrap device of a plurality of nodes constituting the parallel computer system, the bootstrap device is changed according to an instruction from the management terminal device, and a system startup process of a main processor of the plurality of nodes is performed. It is possible to do.

(9) In the management apparatus for a parallel computer system of the above (1), a plurality of the management terminal apparatuses are provided, and means for restricting a function of a part of the plurality of management terminal apparatuses is provided. Things.

The management device for the parallel computer system of (9) includes a plurality of management terminal devices, so that when a specific management terminal device fails, the operation of the parallel computer system is performed by another management terminal device. Perform management.

As described above, when a plurality of management terminal devices are connected to the parallel computer system, the operation contents of the plurality of management terminal devices conflict with the operation contents of other management terminal devices. There are cases.

For this reason, in the management apparatus of the parallel computer system of (9), when the plurality of management terminal apparatuses operate, a specific management terminal apparatus is used as a main management terminal apparatus and another management terminal apparatus is used as a sub-management terminal apparatus. The contention is prevented by setting the management terminal device and limiting the content of the operation performed by the sub management terminal device.

As described above, according to the management apparatus for a parallel computer system, since a plurality of management terminal apparatuses are provided, even if a failure occurs in one management terminal apparatus, the operation management of the parallel computer system is continued. However, it is possible to improve the reliability of the parallel computer system.

Further, according to the management device of the parallel computer system, the main management terminal device and the sub management terminal device are set in the plurality of management terminal devices. Therefore, when the parallel computer system is managed by the plurality of management terminal devices. In addition, it is possible to prevent contention between the operations of the plurality of management terminal devices.

(10) In the management device for a parallel computer system of (1), the management terminal device operates on an auxiliary power supply, and when a specific signal is input, a power-on logic for turning on the main power supply of the management terminal device. And means for transmitting a system control command for turning on the main power supply to the sub-processors of the plurality of nodes collectively or individually when the main power supply is turned on by the power-on logic.

In the management apparatus for a parallel computer system according to the above (10), a power-on logic for turning on the main power supply of the management terminal when the specific signal is input is connected to the management terminal. The power-on logic is connected to a network or another communication line.

At the end of the system startup processing executed when the main power of the management terminal device is turned on,
A program for transmitting a system control command for turning on the main power supply to the sub-processors of the plurality of nodes collectively or individually is added.

Next, a specific signal is sent from another terminal device to the power-on logic via the network or another communication line to turn on the main power of the management terminal device.

When the main power of the management terminal is turned on, a system control command for turning on the main power to the sub-processors of the plurality of nodes collectively or individually after the system startup processing of the management terminal is performed. Is executed, and the operation of the parallel computer system can be started without the operator directly operating the management terminal device.

As described above, according to the management system of the parallel computer system, since the main power of the management terminal device is turned on by access from a remote place, the operation management of the parallel computer system can be performed from a remote place. It is.

[0080]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the drawings together with embodiments.

In all the drawings for describing the embodiments, parts having identical functions are given same symbols and their repeated explanation is omitted.

(Embodiment 1) A schematic configuration of Embodiment 1 for implementing a management device for a parallel computer system according to the present invention will be described below.

FIG. 1 is a diagram showing a schematic configuration of a first embodiment for implementing a management device for a parallel computer system according to the present invention. In FIG. 1, 100a to 100c are nodes, 10
1 is a communication processing device, 102a to 102c are communication interface mechanisms, 103a to 103c are communication cables, 10
4 is a management terminal device, 105a to 105d are system control mechanisms, 106a to 106d are communication cables, 107 is a network concentrator, and 108a to 108d are LAN (L
ocal Area Network) control mechanism, 10
9a to 109d are communication cables, and 110 is a network concentrator.

As shown in FIG. 1, the management apparatus of the parallel computer system according to the present embodiment includes nodes 100a to 100c constituting the parallel computer system and nodes 10 in parallel processing.
Communication processing device 101 for controlling communication in 0a to 100c
And a management terminal device 104 for performing system management of the nodes 100a to 100c, and the management terminal device 104 and the node 10
Network concentrator 10 for connecting 0a-100c
7 and a network concentrator 110. The management terminal device 104 includes a system control mechanism 105a,
The node 100a includes an AN control mechanism 108a, a communication interface mechanism 102a, and a system control mechanism 1
05b and the LAN control mechanism 108b, and the node 1
00b includes a communication interface mechanism 102b, a system control mechanism 105c, and a LAN control mechanism 108c.
c, the system control mechanism 105d, and the LAN control mechanism 1
08d.

As shown in FIG. 1, in the management apparatus of the parallel computer system according to the present embodiment, the nodes 100a to 100a
0c of the communication interface mechanisms 102a to 102c and the communication cables 103a to 103c and the communication processing apparatus 101.
To connect the system control mechanisms 105b to 105d of the nodes 100a to 100c to the communication cables 106a to 106c.
106d and the network control unit 105 connected to the system control mechanism 105a of the management terminal device 104 via the network concentrator 107.
LAN control mechanism 108b of nodes 100a to 100c
108d through the communication cables 109a to 109d and the network concentrator 110 to the L of the management terminal device 104.
It is connected to the AN control mechanism 108a.

As described above, the system control interface of the management device of the parallel computer system according to the present embodiment includes the system control mechanism 105a on the management terminal device 104 side and the system control mechanisms 105b-1 on the nodes 100a-100c side.
The interface is realized by interconnecting the communication device 05d and the network device 05d using a communication cable 106 such as Ethernet (registered trademark) and a network concentrator 107 such as a multi-port repeater.

Further, the system operation support interface of the management device of the parallel computer system according to the present embodiment includes the LAN control mechanism 108 a of the management terminal device 104 and the node 10.
LAN control mechanisms 108b to 108d on the side of 0a to 100c
Are interconnected using a communication cable 109 such as Ethernet and a network concentrator 110 such as a multi-port repeater.

The system operation support interface includes:
This is an interface for managing the operation of the conventional parallel computer system, and is used when the main processors of the nodes 100a to 100c are operating.
The system management is performed by, for example, displaying a message output from the application software executed by the main processor 00c on the management terminal device 104.

Hereinafter, the nodes 100a to 100c constituting the parallel computer system in the management apparatus for the parallel computer system according to the present embodiment will be described.

FIG. 2 is a diagram showing a schematic configuration of the nodes 100a to 100c constituting the parallel computer system in the management apparatus for the parallel computer system according to the present embodiment.

In FIG. 2, reference numeral 200 denotes a main power supply, 201 denotes an auxiliary power supply, 202 denotes a main processor, 203 denotes software, 204 denotes a main memory, 205 denotes a processor memory control mechanism, 206 denotes a system bus, 207 denotes a system disk, and 208 denotes a system disk. I / O control mechanism, 209 is RS-
232C control mechanism, 210 is bootstrap ROM
(Read Only Memory), 211 is a system support mechanism, 212 is a subprocessor, 213 is ROM, 214 is SRAM (Static Lando).
m Access Memory; nonvolatile memory), 2
15 is a local bus, 216 is a power on / off signal, 2
Reference numeral 17 denotes a processor reset signal, 218 denotes a LAN control unit, 219 denotes an RS-232C control unit, 220 denotes a processor, 221 denotes a ROM, and 222 denotes a RAM (Random).
(Access Memory), 223 is a data interface, and 224 is a control interface.

As shown in FIG. 2, the nodes 100a to 100c of the management apparatus of the parallel computer system according to the present embodiment include communication interface mechanisms 102a to 102c, system control mechanisms 105b to 105d, and a LAN control mechanism 108.
b to 108d, a main processor 202 executing application software for performing parallel processing in the nodes 100a to 100c, a system support mechanism 211 having a sub-processor 212, a main power supply 200,
An auxiliary power supply 201 is provided.

The nodes 100a to 100c of the management apparatus of the parallel computer system according to the present embodiment include the operating system executed by the main processor 202 and the software 20 as network software.
3 and main memory 20 for storing software 203
4, a processor memory control mechanism 205 for controlling an interface between the main processor 202 and the main memory 204, a system bus 206, and a system disk 2.
07, an I / O control mechanism 208 for controlling the system disk 207, and an RS-232C control mechanism 20 for outputting node messages and interactively communicating with the operator via the system control mechanisms 105b to 105d.
9 and a bootstrap ROM 210 which stores a bootstrap program for performing a system start-up process of the main processor 202.

In the management apparatus for a parallel computer system according to the present embodiment, the system support unit 211 having a sub-processor 212 and performing system control such as status management of the main processor 202 includes a sub-processor 212.
ROM 21 storing a control program that operates on
3 and SR storing hardware-dependent information
AM 214 is provided.

The system control mechanism 10 of the nodes 100a to 100c of the management device of the parallel computer system according to the present embodiment.
5b to 105d are LAN control units 218 for controlling transmission and reception of Ethernet packets to and from the management terminal device 104.
An RS-232C control unit 219 that controls transmission and reception of RS-232C packets between the RS-232C control mechanism 209 and the sub-processor 212; and a processor 220 that performs protocol conversion between Ethernet packets and RS-232C packets. , A ROM 221 for storing a control program that operates on the processor 220, and a RAM 222 for storing a node message sent from the sub-processor 212 and the RS-232C control mechanism 209.

As shown in FIG. 2, the nodes 100a to 100c of the management device of the parallel computer system according to the present embodiment
The system control mechanisms 105b to 105d are
C control unit 219, data interface 223, and R
S-232C control mechanism 209 and system bus 206
And the main processor 202 via the processor memory controller 205 and the system controller 1
05b to 105d are connected to the sub-processor 212 of the system support unit 211 via the RS-232C control unit 219 and the control interface 224, and the sub-processor 212 of the system support unit 211 is connected to the local bus 215 and the processor memory control unit 205. It is connected to the main processor 202 via the main processor. The sub-processor 212 resets the main processor 202 by a processor reset signal 217 and controls the main power supply 200 by a power on / off signal 216.

As shown in FIG. 2, in the nodes 100a to 100c of the management device of the parallel computer system according to the present embodiment, the system control mechanisms 105b to 105d
-232C control unit 219 and data interface 22
3 and the RS-232C control mechanism 209 are connected to the main processor 202 by connecting the system control mechanisms 105b to 105d and the main processor 202 with a serial interface such as RS-232C. The communication software is compact so that communication between the system control mechanisms 105b to 105d and the main processor 202 is not impaired as much as possible even if a failure occurs in the main processor 202. It is.

The nodes 100a to 100c of the management apparatus of the parallel computer system according to the present embodiment are composed of a part operated by the main power supply 200 and a part operated by the auxiliary power supply 201.

The parts operated by the main power supply 200 include a main processor 202 of the nodes 100a to 100c, a main memory 204 for storing software 203, a processor memory control mechanism 205 for controlling an interface between the main processor 202 and the main memory 204, There are a bootstrap ROM 210 and the like storing a bootstrap program for performing a system start-up process of the main processor 202 of each of the nodes 100a to 100c, and these are connected via the system bus 206 to the communication interface mechanisms 102a to 102c, 1
08b to 108d are connected, and the system disk 207 is connected via the I / O control mechanism 208.

The parts operated by the auxiliary power supply 201 include:
A system support mechanism 211 that controls the main power supply 200 of the nodes 100a to 100c and a system control such as a status management of the main processor 202; and a system control mechanism that controls communication between the nodes 100a to 100c and the management terminal device 104. 105b to 105d.

[0101] The sub-processor 212
The main power supply 200 is controlled by outputting a power on / off signal 216 in accordance with an instruction from the main controller 202 and resetting the main processor 202 by outputting a processor reset signal 217.

The node messages of the nodes 100a to 100c are transmitted to the data interface 223 when the main processor 202 operates and the operating system for controlling the main processor 202 and the software 203 which is network software are activated.
From the RS-232C control mechanism 209 via the RAM 22
2 and the software 203 is not activated, the sub-processor 212 sends a bootstrap message or the like via the control interface 224 to the R.
Stored in AM 222.

The processor 220 of the system control mechanism 105b to 105d of the management apparatus for the parallel computer system according to the present embodiment performs the following processing in addition to the above-described protocol conversion of the packet.

That is, an Ethernet packet from the management terminal device 104 is interpreted, processing is performed in accordance with the contents of the packet, and RAM processing is performed according to an instruction from the management terminal device 104.
The sub-processor 212 performs a process of transmitting the node message stored in the storage device 222 to the management terminal device 104.
Interprets the packet sent via the control interface 224 and performs control according to the contents.

The schematic configuration of the management terminal device 104 of the management device of the parallel computer system according to the present embodiment will be described below.

FIG. 3 is a diagram showing a schematic configuration of the management terminal device 104 of the management device of the parallel computer system according to the present embodiment. In FIG. 3, reference numeral 300 denotes a processor, 301 denotes software, 302 denotes a main memory, 303 denotes a bootstrap ROM, 304 denotes a processor memory control mechanism,
305 is a system bus, 306 is an I / O control mechanism, 30
7 is a system disk, 308 and 309 are RS-232
C control mechanism, 310 is a graphics control mechanism, 311
Is a LAN control unit, 312 is an RS-232C control unit, 31
3 is a processor, 314 is a ROM, 315 is a RAM, 3
Reference numeral 16 denotes a control interface, and 317 denotes a data interface.

As shown in FIG. 3, the management terminal device 104 of the management device of the parallel computer system according to the present embodiment includes a processor 300 for controlling / controlling all processes in the management terminal device 104 and a management terminal device 104. A main memory 302 storing an operating system and software 301 which is network software, and a bootstrap RO storing a bootstrap program for performing a system start-up process of the management terminal device 104.
M303, and a processor memory control mechanism 304 that controls the interface of the processor 300, the main memory 302, and the bootstrap ROM 303.

Further, the management terminal device 104 of the management device of the parallel computer system according to the present embodiment has a system bus 305.
, An I / O control mechanism 306 that controls the system disk 307, a system disk 307, and software 3
RS-23 for issuing a system control command such as power control to the nodes 100a to 100c.
2C control mechanism 308 and RS-232C control mechanism 30 for interactively communicating with the operator via output of node messages and system control mechanism 105a
9, a graphics control mechanism 310 for controlling a man-machine interface such as a display terminal, a keyboard and a mouse, and a system control mechanism 105a.

The system control mechanism 105a of the management terminal device 104 of the management device of the parallel computer system according to the present embodiment.
Is a LAN control unit 311 for controlling transmission and reception of Ethernet packets between the nodes 100a to 100c,
RS-232C between the control mechanisms 308 and 309
RS-232C control unit 312 for controlling transmission and reception of H.232C packets, Ethernet packets and RS-232C
Processor 313 that performs protocol conversion with packets
And a ROM 314 for storing a control program operated by the processor 313, and a RAM 315 for storing node messages sent from the nodes 100a to 100c.

As shown in FIG. 3, the management terminal device 104 of the management device of the parallel computer system according to the present embodiment
The processor 300 is connected to the main memory 302 and the bootstrap ROM 3 via the processor memory control mechanism 304.
03 and the system bus 305, the system disk 307 is connected to the system bus 305 via the I / O control mechanism 306, and the LAN control mechanism 108a and the RS-2
The 32C control mechanisms 308 and 309 and the graphics control mechanism 310 are connected to the system bus 305.

Further, as shown in FIG. 3, in the management terminal device 104 of the management device of the parallel computer system of the present embodiment,
RS-232C control unit 31 of system control mechanism 105a
2 is connected to the RS-232C control mechanisms 308 and 309 via the control interface 316 and the data interface 317.

In the management apparatus for the parallel computer system of the present embodiment, the system control interface
00a to 100c system control mechanisms 105b to 105c
d and the system control mechanism 105a of the management terminal device 104 are interconnected using an Ethernet cable or the like.

The system control interface is in a state where the system control mechanism 105a of the management terminal device 104 is operable, the auxiliary power 201 of the nodes 100a to 100c is turned on, the sub-processor 212 and the system control mechanism 105b to If the node 105d is operable, the main power supply 200 of the nodes 100a to 100c is not turned on, that is, the main processor 202 is not operating, and the operating system and the network software control the entire main processor 202. It can be used even if the software 203 has not been activated.

On the other hand, the system operation support interface is the LAN control mechanism 108a of the management terminal device 104.
And LAN control mechanism 108b of nodes 100a to 100c
And 108d are connected to each other using an Ethernet cable or the like, and the system operation support interface is formed by TCP / IP (Transmi
session Control Protocol / In
For use in Internet Protocol)
Operating systems of the management terminal device 104 and the nodes 100a to 100c, and software 203 and software 301 as network software thereof.
Is activated and can be used only when network software supporting TCP / IP is running.

Hereinafter, the management terminal device 104 of the management device of the parallel computer system according to the present embodiment and the nodes 100a to 100a will be described.
The communication sequence with 00c will be described.

FIG. 4 shows the management terminal device 104 of the management device of the parallel computer system of this embodiment and the nodes 100a to 100a.
It is a figure which shows an example of the communication sequence with 00c. FIG.
, 401 is an adapter control command and its response, 402 is a system control command and its response, and 403 is a node message.

As shown in FIG. 4, in the management apparatus of the parallel computer system according to the present embodiment, transmission and reception of a packet which is an adapter control command and its response 401 or a system control command and its response 402, or a node message 403 The communication is performed by transmitting and receiving.

The adapter control command and its response 401 are transmitted when the software 301 of the management terminal device 104 communicates with the system control mechanism 105a of the management terminal device 104, and when the sub-processor 212 communicates with the system control mechanisms 105b to 105d. Control interface 316 or control interface 224
Sent and received via

The system control command and its response 402 are used when the software 301 of the management terminal device 104 communicates with the sub-processors 212 of the nodes 100a to 100c, and are transmitted and received via the control interface 316 and the control interface 224. .

When the software 203 is not running, the node message 403 indicates that the sub-processor 212
To the RAM 22 of the system control mechanisms 105b to 105d
When the software 203 is activated, the main processor 202
System control mechanism 10 from S-232C control mechanism 209
5b to 105d are transmitted to the RAM 222 and accumulated.

R of system control mechanisms 105b to 105d
The node messages 403 stored in the AM 222 are transmitted from the nodes 100a to
RA of the system control mechanism 105b-105d of 100c
From M222, the RS-232C of the management terminal device 104 is transmitted via the system control mechanism 105a of the management terminal device 104.
The information is transmitted to the control mechanism 309 and displayed on a graphics display or the like connected to the graphics control mechanism 310 of the management terminal device 104.

The packet format of the adapter control command and its response 401 in the management apparatus of the parallel computer system according to the present embodiment will be described below.

FIG. 5 is a diagram showing a packet format of an adapter control command and its response 401 in the management device of the parallel computer system of the present embodiment.
In FIG. 5, reference numeral 501 denotes a type field, 502 denotes a source address field, 503 denotes a destination address field, 504 denotes an information field, and 505 denotes an identifier.

As shown in FIG. 5, a packet of an adapter control command and its response 401 in the management device of the parallel computer system of the present embodiment has a type field 5
01, a source address field 502, a destination address field 503, and an information field 504.
And an identifier 505.

In the management apparatus of the parallel computer system of this embodiment, the type field 501 stores an adapter control command or a packet identifier indicating a response to the adapter control command, for example, “A”. The destination address of the packet is stored in the source address and destination address field 503 of the packet.

In the information field 504, different parameters are stored depending on the type of packet, and an identifier 505 indicating the end of the packet, for example, "LF" (line feed) is added to the end of the packet. You.

Hereinafter, the packet format of the system control command and its response 402 in the management apparatus for the parallel computer system according to the present embodiment will be described.

FIG. 6 is a diagram showing a packet format of a system control command and its response 402 in the management apparatus of the parallel computer system according to the present embodiment.
6, reference numeral 601 denotes a type field, 602 denotes a source address field, 603 denotes a destination address field, 604 denotes an information field, and 605 denotes an identifier.

As shown in FIG. 6, a packet of a system control command and its response 402 in the management device of the parallel computer system of the present embodiment has a type field 6
01, a source address field 602, a destination address field 603, and an information field 604.
And an identifier 605.

In the management apparatus of the parallel computer system of the present embodiment, the type field 601 stores a packet identifier indicating a system control command or its response, for example, “d”, and the source address field 602 stores The source address of the packet and the destination address field 603 store the destination address of the packet.

In the information field 604, different parameters are stored depending on the type of packet, and an identifier 605 indicating the end of the packet, for example, "LF" is added to the end of the packet.

In the management device of the parallel computer system according to the present embodiment, when the hexadecimal “0xffffffff” is stored in the destination address field 603 of the transmission packet from the management terminal device 104, the packet becomes a broadcast packet. , All nodes 1
00a to 100c.

In the management apparatus for the parallel computer system according to the present embodiment, the number added with “0x” indicates a hexadecimal number.

Hereinafter, the system control mechanisms 105a to 105 in the management apparatus for the parallel computer system according to the present embodiment will be described.
The mode transition between the packet mode and the non-packet mode in d will be described.

FIG. 7 is a diagram showing a mode transition of the system control mechanism in the management device of the parallel computer system according to the present embodiment. In FIG. 7, reference numeral 701 denotes a packet mode;
702 is a non-packet mode, and 703 is “SET-MOD
E "command.

As shown in FIG. 7, the system control mechanism in the management apparatus of the parallel computer system according to the present embodiment includes a packet mode 701 for transmitting and receiving fixed-length packets,
A non-packet mode 702 for transmitting and receiving the node message 403 of undefined length is provided. The mode transition between the packet mode 701 and the non-packet mode 702 is performed by an adapter control command “SET-M” from the sub-processor 212.
This is performed by executing the “ODE” command 703.

As described above, the system control mechanisms 105a to 105d of the management device of the parallel computer system according to the present embodiment.
There are two operation modes, a packet mode 701 and a non-packet mode 702.
Is connected to the management terminal device 104 and the plurality of nodes 100a to 100a.
0c is a mode set when performing communication, and the non-packet mode 702 performs connection-type communication with a specific node, and transmits a node message 4 from the specific node.
03 is a mode set when the management terminal 104 is displayed on the management terminal device 104.

In the management device of the parallel computer system according to the present embodiment, the management terminal device 104 and the node 100
a to 100c system control mechanisms 105a to 105d
Operate in the packet mode 701 when the auxiliary power supply 201 is turned on.

Hereinafter, the system control mechanisms 105a to 105 in the management apparatus for the parallel computer system according to the present embodiment will be described.
The transition of the connection state in the non-packet mode 702 of d will be described.

FIG. 8 is a diagram showing the transition of the connection state in the non-packet mode 702 of the system control mechanism in the management apparatus for the parallel computer system according to the present embodiment.
8, 800 is a disconnect state, 801 is a wait connect state, 802 is a connect state, and 803
Is a “SET-CONNECT” command, 804 is the system control mechanism 105a on the management terminal device 104 and the system control mechanisms 105b-1 on the nodes 100a-100c.
05c.

As shown in FIG. 8, in the connection state in the non-packet mode 702 of the system control mechanism in the management apparatus for the parallel computer system of the present embodiment, the other system control mechanism is not connected and the node The disconnect state 800 in which the message 403 is not stored and the node message 403 which is not connected to the partner system control mechanism are stored in the RAM 222.
A wait connect state 801 that is accumulating in the
There is a connect state 802 to which the partner system control mechanism is connected, and the transition of the connection state is “S
ET-CONNECT "command 803 or call control 804 from the system control mechanism.

As shown in FIG. 8, in the management apparatus of the parallel computer system of this embodiment, the non-packet mode 702
At the time of setting, three connection states of a disconnect state 800, a wait connect state 801 and a connect state 802 are held. In the disconnect state 800, communication between the system control mechanisms is disabled. Although communication with the control mechanism is not possible, the node message 403 is stored in the RAM 22
2 are sequentially stored.

When the communicating system control mechanisms are in the connected state 802, connection-type communication in the non-packet mode 702 becomes possible.

These states are described in “SET-CONNECT”.
The transition is made by issuing a “T” command 803. If there is a connection request by the call control 804 from the partner system control mechanism, the wait connection state 80
The state transitions from 1 to the connected state 802.

Table 1 shows a list of adapter control commands and their responses 401 used by the management device of the parallel computer system of this embodiment. In Table 1, the information part indicates information stored in the information part field 504, and the number of byte 0 of the information part is a number indicating the packet type.

[0146]

[Table 1]

Table 2 shows a list of system control commands and their responses 402 used in the management device of the parallel computer system of this embodiment. In Table 2, the information part indicates information stored in the information part field 604, and the number of byte 0 of the information part is a number indicating the packet type.

[0148]

[Table 2]

The processing procedure of the processor 220 of the system control mechanisms 105b to 105d of the nodes 100a to 100c in the management apparatus for the parallel computer system according to the present embodiment will be described below.

FIG. 9 is a flowchart showing a part of the processing procedure of the processor 220 of the system control mechanisms 105b to 105d of the nodes 100a to 100c in the management apparatus for the parallel computer system according to the present embodiment.

As shown in FIG. 9, the nodes 100a to 100c in the management device of the parallel computer system of this embodiment
Processor 2 of the system control mechanism 105b to 105d
In step 20, in the process of step 900, “SET-CO
NNECT "command or call control to
4 to check if there is a connect request.

In the process of step 900, “SET-CO
NNECT "command or call control to
If there is a connect request from step 4, step 901
To check whether the system control mechanisms 105b to 105d of the nodes 100a to 100c are in the non-packet mode 702.

In the process of step 901, the node 100a
If the system control mechanisms 105b to 105d of the nodes 100a to 100c are in the non-packet mode 702, the process proceeds to step 902.
If the system control mechanisms 105b to 105d of the node 00c are not in the non-packet mode 702, the nodes 100a to 100d are processed by the system control command “SET-MODE” from the sub-processor 212 in the process of step 903.
Set the system control mechanisms 105b to 105d of c in the non-packet mode 702, and proceed to the processing of step 902.

In the process of step 902, the node 100
The node messages 403 stored in the RAMs 222 of the system control mechanisms 105b to 105d a to 100c are transmitted to the management terminal device 104 via the system control interface, and the process returns to step 900.

In the processing of step 900, “SET-CO
NNECT "command or call control to
If there is no connection request from the control terminal 4, the process proceeds to step 904, and in the process of step 904, it is checked whether or not there is a system control from the management terminal device 104 by a system control command.

In the processing of step 904, if there is a system control from the management terminal device 104 according to the system control command, the processing proceeds to step 905,
System control mechanism 105b of nodes 100a to 100c
It is checked whether .about.105d is in packet mode 701.

In the process of step 904, if there is no system control from the management terminal device 104 by the system control command, the process proceeds to step 909.

In the processing of step 905, the node 100a
If the system control mechanisms 105b to 105d of the nodes 100a to 100c are in the packet mode 701, the process proceeds to step 906, and the system control mechanism 1 of the nodes 100a to 100c
If 05b to 105d are not the packet mode 701,
In the processing of step 907, the system control command “SET-MODE” from the sub-processor 212
The system control mechanisms 105b to 105d are set to the packet mode 701, and the process proceeds to step 906.

In the process of step 906, the destination address field 603 of the system control command is checked, and the destination address field 603 of the system control command is set to its own logical address or “0xffff”.
If it is "ffffff", the process proceeds to step 908, the contents of the system control command are notified to the sub-processor 212, and the process returns to step 900.

In step 906, if the destination address field 603 of the system control command is not its own logical address and "0xffffffff", the process returns to step 900.

In step 909, the node 100
It is checked whether or not the results of the processing from the sub-processors 212a to 100c have been returned. If the results of the processing have been returned from the sub-processors 212, Step 910 is executed.
Then, a response packet of the system control command is transmitted to the management terminal device 104, and the process returns to step 900.

The processing procedure of the sub-processor 212 of the system support module 211 of the nodes 100a to 100c in the management system of the parallel computer system according to the present embodiment will be described below.

FIG. 10 is a flowchart showing a part of the processing procedure of the sub-processor 212 of the system support mechanism 211 of the nodes 100a to 100c in the management apparatus for the parallel computer system of the present embodiment.

As shown in FIG. 10, nodes 100a to 100a in the management device of the parallel computer system according to the present embodiment.
Sub-processor 21 of system support mechanism 211 of c
In step 2, when the auxiliary power supply 201 is turned on,
In the process of 00, an SRAM for setting the logical addresses of the nodes 100a to 100c and storing a status code to be displayed on a panel provided in the nodes 100a to 100c
“0000” is set in the panel status management area in the area 214.

Next, in the processing of step 1001, the system control mechanisms 105b-1 of the nodes 100a-100c
05d in the non-packet mode 702, and step 1
002, the non-packet mode 70 of the system control mechanisms 105b to 105d of the nodes 100a to 100c.
2 is set to the wait connect state 801.

The mode of the system control mechanisms 105b to 105d of the nodes 100a to 100c is set to the non-packet mode 7
02, and the connection state of the non-packet mode 702 of the system control mechanisms 105b to 105d is set to the wait connect state 801 by the nodes 100a to
The node message of 100c is transmitted to the system control unit 105.
b to 105d, when the connection request is received from the system control mechanism 105a of the management terminal device 104 by the call control 804, the node 10
0a to 100c system control mechanisms 105b to 105d
In order to send the node message stored in the RAM 222 to the management terminal device 104.

Also, by doing so, the nodes 100a-1
It is possible to read the node message stored in the RAM 222 from the management terminal device 104 even when the software 203 on 00c is not activated.

Next, when a system control command is sent from the system control mechanism 105a of the management terminal device 104 to the system control mechanisms 105b to 105d of the nodes 100a to 100c, the system control command is transmitted to the nodes 100a to 100c. Sent to the sub-processor 212,
The sub-processor 212 executes the system control command.

In the processing of step 1003, the system control mechanism 105a of the management terminal device 104
It is checked whether a system control command of a power control instruction for turning on or off the main power supply 200 of the nodes 100a to 100c has been sent via the system control mechanisms 105b to 105d of the devices a to 100c.

In the process of step 1003, the management terminal 1
As a result of checking whether there is a power control instruction from
If there is a power control instruction from the management terminal device 104, the nodes 100a-1
In step 1005, a power control process for turning on or off the main power supply 200 is performed. In step 1005, the execution result of the power control process is reported to the system control mechanisms 105 b to 105 d of the nodes 100 a to 100 c. Return to the processing of.

In the process of step 1003, the management terminal 1
As a result of checking whether there is a power control instruction from
If there is no power control instruction from the management terminal device 104, the process proceeds to step 1006, and the management terminal device 10
4 from the system control mechanism 105a to the nodes 100a to 1
It is checked whether or not a system control command of a panel control instruction for controlling the panels provided in the nodes 100a to 100c has been sent via the system control mechanisms 105b to 105d of 00c.

In the process of step 1006, it is checked whether or not there is a panel control instruction from the management terminal device 104. If there is a panel control instruction from the management terminal device 104, the process proceeds to step 1007. The panel control process is executed, and in step 1008, the result of the panel control process is reported to the system control mechanisms 105b to 105d of the nodes 100a to 100c, and then the process returns to step 1003.

In the process of step 1006, it is checked whether or not there is a panel control instruction from the management terminal device 104. If there is no panel control instruction from the management terminal device 104, the process proceeds to step 1009. From the system control mechanism 105a of the management terminal device 104 to the node 100
It is checked whether a system control command of a reset instruction for resetting the main processor 202 of each of the nodes 100a to 100c has been transmitted via the system control mechanisms 105b to 105d of the nodes a to 100c.

In the process of step 1009, it is checked whether or not there is a reset instruction from the management terminal device 104. If there is a reset instruction from the management terminal device 104, the process proceeds to step 1010, where the node 100a
After the reset process of the main processor 202 of the nodes 100a to 100c is executed and the execution result of the reset process is reported to the system control mechanisms 105b to 105d of the nodes 100a to 100c in the process of Step 1011,
It returns to the process of 03.

As a result of checking in step 1012 whether there is a mode switching request from the system control mechanisms 105b to 105d, if there is a mode switching request, the flow advances to step 1013 to execute the adapter control command "SET". -MODE ", the operation modes of the system control mechanisms 105b to 105d are switched, and the process returns to step 1003.

As described above, according to the management apparatus for the parallel computer system of the present embodiment, the nodes 100a to 100a
0c operating on the auxiliary power supply 201 and the main processor 20
2 uses a network software and communication cables 106b to 106d that are different from the network software and communication cables 109b to 109d used by the communication control device 105.
To the main processor 202 of the plurality of nodes 100a to 100c by transmitting a system control command from the management terminal device 104 to the sub-processor 212 to execute the system control command. Therefore, the operation of the main processor 202 executing the parallel processing and the main processor 20
Irrespective of the operation of the operating system 2 and the software 203 which is network software,
A plurality of nodes 100a to 100a constituting a parallel computer system
The operation management of the management terminal 100c can be collectively performed by the management terminal device 104.

(Embodiment 2) In the management apparatus for a parallel computer system according to the present invention, the management terminal 104 instructs the plurality of nodes 100a to 100d to turn on the main power supply 200, and the status code of the nodes 100a to 100d. Embodiment 2 that monitors whether the main processor 202 of the nodes 100a to 100d has started operation will be described.

FIG. 11 is a diagram showing the configuration of the node 1 from the management terminal device 104 in the management device of the parallel computer system of this embodiment.
FIG. 9 is a diagram showing an example of a power-on sequence instructing 00a to 100d to turn on a main power supply 200. In FIG. 11, reference numeral 100d denotes a node, and reference numerals 1101 to 1112 denote a sequence showing each stage of power-on.

As shown in FIG. 11, in the power-on sequence in which the management terminal device 104 in the management device of the parallel computer system of this embodiment instructs the nodes 100a to 100d to turn on the main power supply 200, the sequence 1101 The auxiliary power supplies 201 of 100a to 100d are turned on.

Auxiliary power supply 20 for nodes 100a-100d
1 is input, the sub-processor 212 of the nodes 100a to 100d initializes the system support mechanism 211 in a sequence 1102, and executes the adapter control command "SET-ADDRESS" to execute the initialization of the system control mechanism 105b to 105d. Initialization and setting of a logical address, which is an address necessary for the management terminal device 104 to manage the nodes 100a to 100d, are performed.

Here, for example, the logical address “0x00”
“SET-ADDRESS” to set “000001”
An example of the format of a command and its response is
It is as follows.

<Command>: A0x00000001: (The destination address field 503 is omitted): 0x01 LF <Response>: A0x00000001: 0x00000001: 0x01 (status information) LF In the sequence 1103, the power of the management terminal device 104 is turned on. Then, the bootstrap program stored in the bootstrap ROM 303 of the management terminal device 104 performs the system startup processing of the management terminal device 104.

In the sequence 1104, the management terminal 1
When the system start-up process of 04 is completed, sequence 1
At 105, the software 301 of the management terminal device 104
Sets the logical address of the management terminal device 104 to “SET-A
DDRESS ”.

The management terminal device 104 and nodes 100a to
When the logical address of 100d is set, the sequence 1
At 106, the software 301 of the management terminal device 104
Reads a status code indicating the status of the nodes 100a to 100d using a broadcast packet of a system control command.

The status codes correspond to nodes 100a-1
00d is managed in the panel status management area in the SRAM 214.
When the auxiliary power supply 201 is turned on normally, a certain status code is written in the panel status management area, and the status code can be read out by the sub-processor 212 (the parallel computer of this embodiment). Code "0000" in the system management device
Can be read. ).

[0186] Here, the management terminal device 104 sets "STA
Using the "TUS-READ" command, the node 100
Broadcast is performed for a to 100d.

The logical address is “0xa00000000”
Is the STATUS-REAA
D ”command is broadcast, and the logical address corresponding to the command is“ 0x00000000 ”.
An example of the format of the response from the node “5” is as follows.

<Command>: d0xa0000000: 0xffffffff: 0x4 LF <Response> d0x00000005: 0xa0000000: 0x04 0000 LF In the sequence 1107, the nodes 100a to 100d receive the "STATUS-READ" command.
The status code “000” is issued by the sub processor 212.
After reading “0”, in sequence 1108, the nodes 100 a to 100 d
A response is returned as described above.

Here, the software 301 of the management terminal device 104 includes the logical address of the node to which a normal response has been returned and the parallel computer stored in advance in the software 301 of the management terminal device 104 or in a specific file. Retry processing is performed by comparing the configuration information of the nodes 100a to 100d constituting the system with each other and transmitting a "STATUS-READ" command again at predetermined time intervals to nodes that do not return a normal response. Do.

In sequence 1109, management terminal device 1
04, the main power supply 2 of the node from which the normal response packet was returned in the sequence 1108
00 is input by a “P-ON” command.

For example, if the logical address is “0xa0000”
000 ”from the management terminal device 104 to the node whose logical address is“ 0x00000005 ”.
An example of the format of the “-ON” command and its response is as follows.

<Command>: d0xa0000000: 0x00000005: 0x01LF <Response>: d0x00000005: 0xa0000000: 0x01 (completion code) LF At this time, under the control of the software 301 of the management terminal device 104, " PO
N "while shifting the command
The inrush current to the power supply equipment that supplies power to the entire parallel computer system can be reduced.

In sequence 1110, the sub-processors 212 of the nodes 100a to 100d that have received the "P-ON" command output the power-on signal 216, turn on the main power supply 200, and then execute the "P-ON" command. Is returned to the management terminal device 104.

Main power supply 200 for nodes 100a-100d
Is turned on, the main processor 202 executes the bootstrap program stored in the bootstrap ROM 210, and the system startup processing is started.

The status codes incremented by the bootstrap program during the system start-up processing include, for example, the following. Here, in the management device of the parallel computer system of the present embodiment, the status code is indicated by a hexadecimal number.

[0196]

[Table 3]

The bootstrap program of the main processor 202 of the nodes 100a to 100d is
The status code is written into the panel status management area in the SRAM 214 of 00a to 100d, and when the system start-up process proceeds, the status code is periodically updated.

Further, the panel status management area includes:
It can be referred to from the sub-processors 212 of the nodes 100a to 100d, for example, the nodes 100a to 100d.
The status code can be disclosed to the operator by displaying the status code on a display device such as a panel provided in the device.

Software 301 of management terminal device 104
Periodically reads the status codes of the nodes 100a to 100d by using a “STATUS-READ” command, thereby obtaining the nodes 100a to 100d.
Monitor the state of d.

In sequence 1110, management terminal device 1
The software 301 of 04 monitors a certain period of time from transmission of the system control command to reception of its response. As shown in FIG. 11, some failure has occurred, and a normal response has occurred within a certain period of time. For the node 100d that does not return, a retry process of transmitting a system control command again at a predetermined time interval is performed in sequence 1111.

As shown in FIG. 11, in the management device of the parallel computer system of the present embodiment, a normal response is not returned from the node 100d as a result of the retry process of a fixed number of times (three times in the present embodiment). In this case, in the sequence 1112, the software 30 of the management terminal
1 recognizes that a failure has occurred in the node 100d.

The software 301 of the management terminal device 104
As described above, when a normal response to a specific system control command is not received within a predetermined time, control is performed to transmit the specific system control command again at a predetermined time interval. Thus, even if the software 203 of the nodes 100a to 100d is not activated, it is possible to manage whether or not the system startup processing of the main processor 202 of the nodes 100a to 100d has been completed normally.

As described above, according to the management apparatus of the parallel computer system of the present embodiment, the main power supply 200 of the plurality of nodes 100a to 100d is turned on or off according to an instruction from the management terminal device 104. The main power supply 200 of the plurality of nodes 100a to 100d constituting the computer system can be turned on or off by the management terminal device 104 collectively or individually.

Further, according to the management apparatus for the parallel computer system of the present embodiment, the instruction to turn on the main power supply 200 to the nodes 100a to 100d is issued at a specific time interval set in advance. It is possible to suppress the inrush current of the power supply equipment that supplies the power.

Further, according to the management device of the parallel computer system of the present embodiment, the status codes of the nodes 100a to 100d are read in accordance with an instruction from the management terminal device 104, so that the status of the plurality of nodes 100a to 100d can be determined. It is possible to manage them collectively at 104.

Further, according to the management apparatus for a parallel computer system of the present embodiment, it is checked whether or not a normal response to a specific system control command from the management terminal apparatus 104 is received within a predetermined time. The management terminal device 104 can monitor whether a plurality of nodes constituting the system are operating normally.

(Embodiment 3) In the following, a description will be given of a third embodiment in which a node message 403 from the node 100a is displayed on the management terminal device 104 and maintenance is performed as necessary in the management device of the parallel computer system of the present invention. I do.

FIG. 12 shows a case where the node 10 is added to the management terminal device 104 in the management device of the parallel computer system of this embodiment.
It is a figure showing an example of a sequence which displays node message 403 from 0a. In FIG.
Reference numeral 1217 denotes a sequence of each stage for displaying the node message 403.

As shown in FIG. 12, in the sequence of displaying the node message 403 from the node 100a on the management terminal device 104 in the management device of the parallel computer system of this embodiment, in the sequence 1201, the node 100a Power supply 201 is turned on,
System control mechanism 105b (operation mode is packet mode 701), sub-processor 212 and processor 22
0 is operable.

Node 1 to which auxiliary power supply 201 is turned on
00a sub-processor 212
In the "SET-ADDRESS" command, node 1
00a logical address is set.

Next, the sub processor 2 of the node 100a
12 is a sequence 1203, “SET-MODE”
The operation mode of the system control mechanism 105b is changed to a non-packet mode 702 by a command (disconnect state 800).
Set to.

Sub processor 212 of node 100a
Is a sequence 1204, and “SET-CONN
The ECT command sets the connection state to the wait connect state 801 in the non-packet mode 702.

On the other hand, when the power of the management terminal device 104 is turned on in the sequence 1205, the management terminal device 104 starts the system start-up process of the management terminal device 104.

When the system start-up process of the management terminal device 104 is completed, in a sequence 1206, the software 301 of the management terminal device 104 uses the “SET-ADDRESS” command to execute the The logical address is set, and the operation mode is set to the disconnect state 800 of the non-packet mode 702 by using “SET-MODE” in sequence 1207.

In sequence 1208, management terminal device 1
04 software 301 is “STATUS-REA
When the status code of the node 100a is read by the “D” command and the status code “0000” can be read, a “P-ON” command is transmitted to the node 100a in a sequence 1209 to instruct the main power supply 200 of the node 100a to be turned on. I do.

The node 100 that has received the “P-ON” command from the management terminal device 104 and turned on the main power supply 200
a stores a bootstrap program stored in a bootstrap ROM 210 in the main processor 202;
To perform the system startup processing of the node 100a.

At this time, node message 403 output from the bootstrap program of node 100a
Are stored in the RAM 222 of the system control mechanism 105b of the node 100a via the sub-processor 212.

Software 301 of management terminal device 104
In the sequence 1210, “SET-CONNECT
The “T” command changes the connection state of the system control mechanism 105 a of the management terminal device 104 to the connected state 80.
2, the system control mechanism 1 of the node 100a
05a stored in the RAM 222 of the node 100a
Of the node message 403 during the system start-up process of the main processor 202 is started.

The system control unit 105a of the management terminal device 104 having received the “SET-CONNECT” executes the sequence 1211 in the system control unit 1 of the node 100a.
05b and the call control 804, and the node 10 receiving the call control 804
0a, the connection state of the system control mechanism 105b is changed from the wait connect state 801 to the connect state 80a.
Transition to 2.

At the same time, the system control mechanism 105b of the node 100a executes “REPORT-
A "CONNECT" command is issued to the sub-processor 212 of the node 100a to notify that there is a connect request from the management terminal device 104.

At this time, "REPORT-CONNECT"
An example of the format of the "T" command and its response is as follows. In addition, the following "REPORT-C
In the “ONNECT” command and its response, the transmission / reception address is omitted.

<Command>: A :: 0xA (Connect state change status) LF <Response>: A :: 0xA LF In the sequence 1213, the node message 403 of the node 100a is transmitted to the system control mechanism 1 of the node 100a.
When the connection request 05b receives the connect request by the call control 804, the connection request is transmitted from the RAM 222 of the system control mechanism 105b of the node 100a to the management terminal device 104 via the LAN control unit 218.

Although not specifically shown in FIG. 12,
If an inconsistency is found in the file system used by the main processor 202 of the node 100a during the system start-up process of the main processor 202 of the node 100a and the system start-up process is interrupted, in a sequence 1214, The operator operates the management terminal device 1
Based on the node message 403 during the system start-up process of the main processor 202 of the node 100a output to the display device 04, it is recognized that a failure has occurred in the node 100a, and a file system such as UNIX fsck is inspected. By inputting a maintenance command, the node 100a can be maintained.

Although the system start-up process of the main processor 202 of the node 100a has been completed normally, the main processor
If the main processor 202 of the node 100a outputs a panic message and the system goes down while using the node 02, the operator sends the node message 403 stored in the RAM 222 of the system control mechanism 105b of the node 100a to the management terminal. It is also possible to display on the device 104 and examine the cause of the system down based on the content of the node message 403.

When disconnecting the connection with the node 100a, the software 301 of the management terminal device 104
In sequence 1215, "SET-CONNECT"
Issue a command.

Software 301 of management terminal device 104
Issues a “SET-CONNECT” command, thereby causing the system control mechanism 105 of the management terminal device 104 to issue a “SET-CONNECT” command.
a goes to the disconnect state 800 and the sequence 12
At 16, the system control mechanism 10 of the management terminal device 104
5a performs call control 804 on the system control mechanism 105b of the node 100a.

At sequence 1217, node 100a
The system control mechanism 105b of the management terminal device 1
04, the call control 804 from the system control mechanism 105a recognizes that a connection disconnection request has been made from the partner, and at the same time reports this to the sub-processor 212 of the node 100a by a "REPORT-CONNECT" command. .

As described above, by transmitting a system control command from the management terminal device 104 to the nodes 100a to 100c, the management terminal device 104
Power control of main power supply 200 and nodes 10a to 100c
Status monitoring of 0a to 100c becomes possible.

As described above, according to the management apparatus of the parallel computer system of the present embodiment, the packet mode 701
And the non-packet mode 702 is switched by the sub-processor 212, so that a system control command and its response can be
00a to 100c and communication for setting a connection with a specific node and continuously receiving a node message 403 of the specific node are the same without using special hardware such as a switching device. Can be performed by the management terminal device 104.

Further, according to the management apparatus for a parallel computer system of the present embodiment, the main processor 2 of a specific node
02 or the sub-processor 212 accumulates the node message 403 output during operation, and the management terminal device 104 reads the accumulated node message 403. Therefore, a plurality of nodes 100 constituting the parallel computer system
Even if the main processors 202 a to 100 c output the node message 403 and stop the operation, the node message 403 is transmitted to the management terminal device 104.
Can be managed collectively.

(Embodiment 4) Hereinafter, in the management apparatus for a parallel computer system according to the present invention, the nodes 100a to 100a
A fourth embodiment will be described in which the system start-up process of the main processor 202 is executed and the bootstrap device is changed as necessary to perform the system start-up process.

In the management apparatus of the parallel computer system according to the present embodiment, when the load processing from the bootstrap device ends abnormally, the contents are changed by reading and writing the contents of the main memory 204 of the nodes 100a to 100c. By resetting the main processor 202 of each of the nodes 100a to 100c, a load process from another bootstrap device can be performed.

In the management apparatus of the parallel computer system according to the present embodiment, when reading and writing the contents of the main memory 204 of the nodes 100a to 100c, "MS-REA"
D and MS-WRITE commands. These system control commands are sent to node 100
It can also be used to investigate the contents of the main memory 204 of the nodes 100a to 100c when a failure occurs during the execution of parallel processing as a normal task in the main processor 202 of the nodes 100a to 100c.

In the management apparatus of the parallel computer system according to the present embodiment, the contents of the registers in the hardware modules such as the main processor 202, the sub-processor 212, and the processor memory control mechanism 205 of the nodes 100a to 100c can be read and written. In that case, it is possible to use the “REG-READ” command and the “REG-READ” command.
WRITE "command.

For example, if a failure occurs in the main processor 202 of each of the nodes 100a to 100c during the execution of parallel processing, which is a normal operation, the nodes 100a to 100c
The failure log collected by each hardware resource in the register 0c is saved in a register, and the above-mentioned “REG-READ”
By reading the failure log from the management terminal device 104 using a command, it is possible to specify the cause of the failure from the management terminal device 104.

In the management apparatus of the parallel computer system according to the present embodiment, the status code of the nodes 100a to 100c is read by the "STATUS-READ" command, and if there is a node in which the system is down, "PROC-RESET" By transmitting the “.” Command, an operation of resetting and restarting the main processor 202 of the node in which the system is down can also be performed.

Hereinafter, the system startup processing of the main processor 202 of each of the nodes 100a to 100c in the management apparatus for the parallel computer system according to the present embodiment will be described.

FIG. 13 is a flowchart showing a system start-up procedure of the main processor 202 of each of the nodes 100a to 100c in the management apparatus for the parallel computer system according to the present embodiment.

FIG. 14 shows the SR of the nodes 100a to 100c in the management device of the parallel computer system according to the present embodiment.
It is a figure showing the memory map of AM214. 14, reference numeral 1400 denotes primary bootstrap path information, and 1401 denotes alternate bootstrap path information.

As shown in FIG. 14, the nodes 100a to 100 in the management apparatus of the parallel computer system according to the present embodiment.
The memory map of the SRAM 214c stores the first software 203 storing the software 203 such as the operating system.
And the alternative bootstrap path information 1401 indicating the bootstrap device to be used when the first bootstrap device cannot be used.

As shown in FIG. 13, the nodes 100a to 100a in the management apparatus of the parallel computer system according to the present embodiment.
In the system start-up processing procedure of the main processor 202c, when the main power supply 200 is turned on in response to a power-on instruction from the management terminal device 104, the main processor 202
The bootstrap program stored in the bootstrap ROM 210 is executed by the
In the process of 00, the status code of the panel status management area in the SRAM 214 is set to “1000”, and in the process of step 1301, the initial diagnosis and initialization of each hardware module in the nodes 100a to 100c are performed.

In the processing of step 1302, step 1
It is checked whether or not the initial diagnosis and initialization of the hardware in the processing of 301 have been completed normally. If the initial diagnosis and initialization of the hardware have been normally completed in the processing of step 1301, the processing of step 1303 is performed. move on.

If the initial diagnosis and initialization of the hardware have been abnormally terminated in the processing of step 1301, the status code "1FFF" is displayed on the panel provided for the node in the processing of step 1313, The system startup process of the main processor 202 of the node ends abnormally.

In the process of step 1303, the status code is set to “2000”, and in the process of step 1304, the hardware-dependent information stored in the SRAM 214 is referred to the bootstrap path information shown in FIG. Bootstrap path information 1400
The software 203 such as an operating system is loaded into the main memory 204 from a bootstrap device (for example, the system disk 207 or the like) specified by the user.

The bootstrap path information in the SRAM 214 is copied to a specific area in the main memory 204 when a bootstrap program stored in the bootstrap ROM 210 is executed, and can be referred to by the software 203 when the system starts up.

In the management apparatus for the parallel computer system according to the present embodiment, the bootstrap device includes a base file of a bootp protocol (Request For Connect boot) via an Ethernet (system control interface) in addition to a local file in the own node. A bootstrap file that can be obtained from another node connected to the Ethernet using the same protocol is also applicable.

In the processing of step 1305, it is checked whether or not the load processing for loading the software 203 such as the operating system from the bootstrap device designated by the primary bootstrap path information 1400 into the main memory 204 has been completed normally. If the loading process is successful, step 1306
If it fails, the process proceeds to step 1314.

In the process of step 1306, the status code is set to “3000”, and the software 203 loaded in the main memory 204 is started.
In step 07, the status code is set to "A000". In step 1308, various system parameters are set. In step 1309, the file system is initialized. In step 1310, TCP is set.
Initialize the network such as / IP.

In the management apparatus for the parallel computer system of the present embodiment, the system operation support interface using the functions of the software 203 such as the operating system and network software operated by the main processor 202 of the nodes 100a to 100c is used at this time. It becomes possible.

In step 1311, the application software is started. In step 1312, the status code is set to “F000”, and the system start-up process of the main processor 202 ends.

On the other hand, in the process of step 1314, the SR
The bootstrap device specified in the alternate bootstrap path information 1401 is referred to by referring to the alternate bootstrap path information 1401 in the AM 214 (the DAT (Digital Audio) is not particularly disclosed in the management apparatus of the parallel computer system of the present embodiment.
o) and software 203 such as an operating system from the input / output device (e.g., Tape).
Load 4

In the processing of step 1315, it is checked whether or not loading from the bootstrap device specified by the alternate bootstrap path information 1401 has been successful. If the loading has succeeded, the flow proceeds to the processing of step 1306.

If the loading from the bootstrap device specified by the alternate bootstrap path information 1401 does not succeed in the processing of step 1315, the loading processing is performed by the operator by specifying the bootstrap device in the processing of step 1316. .

In the processing of step 1317, step 1
It is checked whether or not the loading process by the operator by specifying the bootstrap device in the process of 316 has been completed normally. If the loading process has been completed normally, the process proceeds to step 1306. If not, the process proceeds to step 1306.
In the process of 318, the status code is set to “2FFF”, and the system startup process of the main processor 202 ends abnormally.

The nodes 100a to 100a-1 performed as described above
If the system startup process of the main processor 202 at 00c ends abnormally, the system startup process with a changed bootstrap device is further performed as follows.

Software 301 of management terminal device 104
Are connected to the nodes 100a to 100a via the system control interface using the "MS-READ" command.
The bootstrap device in which the main processor 202 has failed in the system start-up process is checked by referring to the specific area in the main memory 204 of FIG.

Next, the software 301 of the management terminal device 104 uses the “MS-WRITE” command and sends the information to the nodes 100a to 100d via the system control interface.
A bootstrap device name other than the bootstrap device for which the main processor 202 failed to start up the system is written in the specific area of the main memory 204 of the main memory 204 where the bootstrap path information is stored.

The software 301 of the management terminal device 104
Resets the main processor 202 of the nodes 100a to 100c using the "PROC-RESET" command after rewriting the bootstrap path information of the specific area of the main memory 204 of the nodes 100a to 100c as described above. Then, the system startup process of the main processor 202 is performed again, whereby the system startup process in which the bootstrap destination is changed can be performed.

Rewriting of the bootstrap path information can also be performed by the following method when the system startup processing of the main processor 202 of each of the nodes 100a to 100c has been completed normally.

That is, S of the nodes 100a to 100c
The bootstrap path information of the RAM 214 is stored in the node 1
Since the software 203 of the management terminal device 104 can be rewritten from the software 203 of the management terminal device 104,
01 is transmitted to the software 203 of the nodes 100a to 100c via the system operation support interface.
An instruction is given to rewrite the bootstrap path information, and the instructed software 203 rewrites the bootstrap path information of the node.

The software 203 of the nodes 100a to 100c notifies the updated bootstrap path information to the software 301 of the management terminal device 104 via the system control interface, and the software 301 of the management terminal device 104 transmits the updated bootstrap path information via the system control interface. The main processor 202 of each of the nodes 100a to 100c using the “PROC-RESET” command
Is reset, the load process from the updated bootstrap path is performed immediately.

As described above, according to the management device of the parallel computer system of the present embodiment, the contents of the main memory 204 or the registers of the nodes 100a to 100c are referred to or updated by the instruction from the management terminal device 104. A plurality of nodes 100 constituting a parallel computer system
The contents of the main memory 204 at the time of occurrence of a failure in a to 100c can be collectively managed by the management terminal device 104.

Further, according to the management apparatus of the parallel computer system of the present embodiment, the main processor 202 of each of the nodes 100a to 100c is instructed by the management terminal device 104.
Of the main processor 2 of the plurality of nodes 100a to 100c constituting the parallel computer system.
02 can be collectively performed from the management terminal device 104.

According to the management apparatus for a parallel computer system of the present embodiment, the management terminal device 104
0a to 100c, the bootstrap path information in the main memory 204 of the nodes 100a to 100c can be changed according to an instruction from the management terminal device 104, and the main processor 20
2 is performed, when a failure occurs in a specific bootstrap device of a plurality of nodes 100a to 100c configuring the parallel computer system, the bootstrap device is changed according to an instruction from the management terminal device 104 to change the node. It is possible to perform the system start-up processing of the main processors 202 of 100a to 100c.

(Embodiment 5) The following describes a schematic configuration of Embodiment 5 in which the reliability of a management device for a parallel computer system of the present invention is improved using a plurality of management terminal devices.

FIG. 15 is a diagram showing a schematic configuration of the fifth embodiment in which the management terminal device is duplicated in the management device of the parallel computer system of the present invention. In FIG.
e is a system control mechanism, 106e is a communication cable, 10
8e is a LAN control mechanism, 109e is a communication cable, 11
Reference numeral 1 denotes a management terminal device.

As shown in FIG. 15, the management device of the parallel computer system according to the present embodiment includes a management terminal device 111, a communication cable 106e, and a communication cable 109e.
The management terminal device 111 includes a system control mechanism 105e,
And a system control mechanism 105e of the management terminal device 111.
6e, it is connected to the network concentrator 107, and the LAN control mechanism 108e of the management terminal device 111 is connected to the network concentrator 110 via the communication cable 109e.

As described above, in the management apparatus of the parallel computer system according to the present embodiment, a plurality of management terminal apparatuses 104 and 1
11, the management of the parallel computer system can be continued by another management terminal even if one management terminal fails. When the operation management of the parallel computer system is performed, the contents of the system control command and the adapter control command transmitted by the plurality of management terminal devices may conflict with each other. It is necessary to perform a process for preventing contention of operation of the device.

In the following, a description will be given of a processing procedure for preventing contention between operations of the management terminal devices when a plurality of management terminal devices are used in the management device of the parallel computer system of the present embodiment.

FIG. 16 is a flowchart showing a processing procedure for preventing a conflict between the operations of a plurality of management terminal devices in the management device of the parallel computer system according to the present embodiment.

In the management device of the parallel computer system of the present embodiment, when the management terminal device is duplicated,
The duplication information of the management terminal device is stored in, for example, the management terminal device 1.
04 and the software 30 of the management terminal device 111
By setting in advance to a storage area that can be referenced from 1,
It is possible to prevent contention between duplicated management terminal devices.

As shown in FIG. 16, in the processing procedure when the management terminal device is duplicated in the management device of the parallel computer system of the present embodiment, the management terminal device 104 and the management terminal device 111 are processed in step 1600. Both software 301 refer to the duplication bit indicating that the management terminal device is duplicated, and when the bit is set, recognize that the management terminal device is duplicated.

In the processing of step 1601, the IP address of the partner management terminal device is obtained via a network (for example, a system operation support interface).

In the process of step 1602, in order to determine the main management terminal device and the sub-management terminal device for backing up the main management terminal device, for example, the one with the smaller IP address is determined as the main management terminal device, A sub-management terminal device.

At this time, only the main management terminal is operated, and when a failure occurs in the main management terminal, an operation method for immediately switching to the sub management terminal is described. And the operation method of operating both at the same time, but in the latter case, the contents of the system control command for controlling the nodes 100a to 100c and the contents of the adapter control command from both sides may conflict with each other. The system control command and the adapter control command that can be transmitted from the sub-management terminal device are partially restricted.

For example, in the process of step 1603, it is determined whether or not the self-management terminal device is the main management terminal device. If the self-management terminal device is not the main management terminal device, the system control command is executed in the process of step 1604. ("P-
ON, P-OFF, etc.) and the adapter control command (“SET-CONNECT”, etc.) are prohibited from being issued, so that the contents of the system control command for controlling the nodes 100a to 100c and the adapter control command can be changed. Even if there is a conflict, it is possible to maintain consistency as a parallel computer system.

As described above, according to the management apparatus for a parallel computer system of the present embodiment, since a plurality of management terminal apparatuses are provided, even if a failure occurs in one management terminal apparatus, the parallel computer system can be used. It is possible to continue operation management and improve the reliability of the parallel computer system.

Further, according to the management apparatus for a parallel computer system of the present embodiment, the main management terminal apparatus and the sub management terminal apparatus are set in the plurality of management terminal apparatuses, so that the parallel computer system can be managed by the plurality of management terminal apparatuses. When the management is performed, it is possible to prevent a conflict between the operations of the plurality of management terminal devices.

(Embodiment 6) In the management apparatus for a parallel computer system according to the present invention, a power-on logic operating with an auxiliary power supply is added to the management terminal apparatus 104, and the main power supply of the management terminal apparatus 104 is changed from a remote place. A sixth embodiment in which the main power of the parallel computer system is turned on by turning on the power will be described.

FIG. 17 is a diagram showing a schematic configuration of hardware in the management terminal device when a power-on logic operating on an auxiliary power supply is added to the management terminal device 104 in the management device of the parallel computer system of the present embodiment. is there. In FIG. 17, 1700 is an auxiliary power supply, 1701 is a power-on logic,
1702 is a power supply control signal, 1703 is a main power supply, 1704
Is a terminal device, and 1705 is a network.

As shown in FIG. 17, the management terminal device 104 in the management device of the parallel computer system according to the present embodiment comprises:
An auxiliary power supply 1700, a power-on logic 1701, and a main power supply 1703 are provided.
2 and to another terminal device 1704 via the network 1705.

As shown in FIG. 17, the management terminal device 104 in the management device of the parallel computer system according to the present embodiment comprises:
A power-on logic 1701 that operates on the auxiliary power supply 1700 is provided. The power-on logic 1701 includes a network control unit, a power control unit, a microprocessor, and the like (not shown). When the power supply control instruction is received by the
03 is provided.

The power-on logic 1701 enables the following remote operation of the management terminal device 104, for example.

In the management apparatus for the parallel computer system of this embodiment, another terminal device 1704 connected via the network 1705 logs in to the power-on logic 1701 using, for example, the telnet protocol. (At this time, the auxiliary power supply 1700 has been turned on to the management terminal device 104.) Next, the terminal device 1704 issues a power-on command to the power-on logic 1701.
Upon receiving the power-on command, the power-on logic 1701 recognizes that an external power-on instruction has been issued, outputs a power control signal 1702, and turns on the main power 1703 of the management terminal device 104.

When the main power supply 1703 of the management terminal device 104 is turned on, the bootstrap program stored in the bootstrap ROM 303 stores the bootstrap program in the management terminal device 10.
4 performs the system start-up processing, and the software 301
Start

FIG. 18 is a diagram showing an example of a shell program of the management terminal device 104 in the management device of the parallel computer system according to the present embodiment. Here, the shell program is UNI, which is a general-purpose operating system.
X refers to a program that performs a series of procedures and describes a plurality of command names or program names to be executed in X. In FIG. 18, instead of UNIX command names or program names, functions of the commands are simply described. are doing.

As shown in FIG. 18, the shell program executed at the time of the system start-up process of the management terminal device 104 of the management device of the parallel computer system according to the present embodiment includes in advance the main power supply 200 of the nodes 100a to 100c. A system control command for instructing the management terminal device 104 to be turned on is described, and this shell program is executed when the main power supply 1703 of the management terminal device 104 is turned on.

By doing so, the management terminal device 104
, The main power supply 200 of the nodes 100a to 100c is turned on, and the system start-up process of the main processor 202 of the nodes 100a to 100c can be automatically performed.

As described above, according to the management apparatus for a parallel computer system of the present embodiment, the main power supply 1703 of the management terminal device 104 is turned on by access from a remote place,
Further, activation of the nodes 100a to 100c (main power supply 200
), And operation and management of the parallel computer system can be performed from a remote place.

The management apparatus for a parallel computer system according to the present embodiment described above is not particularly shown, but can be applied to the following system.

(1) In a dedicated parallel computer system which does not have a general-purpose operating system on each node and executes a specific function, only the system control interface which does not use the network function of the general-purpose operating system is used. Perform operation management.

According to the management apparatus for a parallel computer system of the present invention, the system control interface has an auxiliary power supply and a network function that are independent of the processor to be managed and managed. Since a network function such as an IP is not used, the present invention can be applied to a dedicated parallel computer system not equipped with the general-purpose operating system.

(2) In a parallel computer system in which each node does not have a function of operating with an auxiliary power supply or has a limited function of operating with an auxiliary power supply, only the system operation support interface which does not use an auxiliary power supply Is managed using.

In this case, the function of the system control interface is substituted by the system operation support interface except for the function that requires an auxiliary power supply such as turning on the main power supply, so that the parallel computer system management device of the present invention can be used. It is possible to apply.

As described above, the present invention has been specifically described based on the above-described embodiment. However, the present invention is not limited to the above-described embodiment, and may be variously modified without departing from the gist thereof. Of course.

[0296]

The effects obtained by the representative ones of the inventions disclosed in the present application will be briefly described as follows.

(1) Operate with auxiliary power supplies of a plurality of nodes,
Transmitting a system control command from the management terminal device to a system control mechanism that communicates with the management terminal device using network software and a communication cable different from the network software and communication cable used by the main processor; Since the main processor of a plurality of nodes is controlled by executing the control command in the sub-processor operated by the auxiliary power supply,
Regardless of the operation of the main processor executing the parallel processing and the operation of the operating system and network software of the main processor, the operation management of a plurality of nodes constituting the parallel computer system can be collectively performed by the management terminal device. It is.

(2) Since the main power of a plurality of nodes is turned on or off in accordance with an instruction from the management terminal device, the turning on or off of a plurality of nodes constituting the parallel computer system is collectively or individually performed by the management terminal device. It is possible to do.

(3) Since an instruction to turn on the main power supply to a plurality of nodes is given at a predetermined specific time interval, the rush current of the power supply equipment for supplying power to the parallel computer system can be suppressed low. is there.

(4) It is checked whether or not a normal response to a specific system control command from the management terminal device is received within a predetermined time, so that a plurality of nodes constituting the parallel computer system are operating normally. Can be monitored by the management terminal device.

(5) Since the main processor or the sub-processor of a specific node accumulates node messages output during operation and the management terminal reads the accumulated node messages, a plurality of nodes constituting the parallel computer system are stored. Even when the main processor outputs the node message and stops its operation, the node message can be collectively managed by the management terminal device.

(6) Since the contents of the main memory or the register of the node are referred to or updated according to the instruction from the management terminal device, the contents of the main memory and the register when a failure occurs in a plurality of nodes constituting the parallel computer system are managed. It can be managed collectively by the terminal device.

(7) Since the main processors of a plurality of nodes are reset in accordance with an instruction from the management terminal device, the reset of the main processors of a plurality of nodes constituting the parallel computer system can be collectively performed from the management terminal device. It is possible.

(8) Since the bootstrap path information in the main memory of a plurality of nodes is changed and the main processor is reset according to an instruction from the management terminal device, a specific boot of a plurality of nodes constituting the parallel computer system is performed. When a failure occurs in the strap device, it is possible to change the bootstrap device and perform system startup processing of the main processors of the plurality of nodes according to an instruction from the management terminal device.

(9) Since a plurality of management terminal devices can be provided, even if a failure occurs in one management terminal device, the operation management of the parallel computer system can be continued, and a plurality of management terminal devices can be managed. Since the terminal device is set as the main management terminal device and the sub-management terminal device, when the parallel computer system is managed by a plurality of management terminal devices, it is possible to prevent a conflict between the operations of the plurality of management terminal devices. is there.

(10) Since the main power of the management terminal device is turned on and the main power of a plurality of nodes is turned on by access from a remote place, the operation management of the parallel computer system can be performed from a remote place. .

[Brief description of the drawings]

FIG. 1 is a diagram showing a schematic configuration of a first embodiment for implementing a management device of a parallel computer system of the present invention.

FIG. 2 is a diagram illustrating a schematic configuration of a node configuring the parallel computer system in the management device of the parallel computer system according to the first embodiment.

FIG. 3 is a diagram illustrating a schematic configuration of a management terminal device in the management device of the parallel computer system according to the first embodiment.

FIG. 4 is a diagram illustrating an example of a communication sequence between a management terminal device and each node in the management device of the parallel computer system according to the first embodiment.

FIG. 5 is a diagram illustrating a packet format of an adapter control command and its response in the management device of the parallel computer system according to the first embodiment.

FIG. 6 is a diagram illustrating a packet format of a system control command and its response in the management device of the parallel computer system according to the first embodiment.

FIG. 7 is a diagram illustrating a mode transition of a system control mechanism in the management device of the parallel computer system according to the first embodiment.

FIG. 8 is a diagram showing transition of a connection state in a non-packet mode of a system control mechanism in the management device of the parallel computer system according to the first embodiment.

FIG. 9 is a flowchart illustrating a part of a processing procedure of a processor of a system control mechanism in the management device of the parallel computer system according to the first embodiment.

FIG. 10 is a flowchart illustrating a part of a processing procedure of a sub-processor of a system support mechanism in the management device of the parallel computer system according to the first embodiment.

FIG. 11 is a diagram showing an example of a power-on sequence instructing each node to turn on main power from a management terminal device in the management device of the parallel computer system according to the second embodiment.

FIG. 12 is a diagram illustrating an example of a sequence for displaying a node message of each node on a management terminal device in the management device of the parallel computer system according to the third embodiment.

FIG. 13 is a flowchart illustrating a system startup processing procedure of a main processor of a node in the management device for the parallel computer system according to the fourth embodiment.

FIG. 14 is a diagram illustrating a memory map in an SRAM in a node in the management device of the parallel computer system according to the fourth embodiment.

FIG. 15 is a diagram illustrating a schematic configuration of a fifth embodiment in which a management terminal device is duplicated in the management device of the parallel computer system of the present invention.

FIG. 16 is a flowchart illustrating a processing procedure for preventing contention between operations of a plurality of management terminal devices in the management device of the parallel computer system according to the fifth embodiment.

FIG. 17 is a diagram illustrating a schematic configuration of hardware in the management terminal device when a power-on logic operating on an auxiliary power supply is added to the management terminal device in the management device of the parallel computer system according to the sixth embodiment.

FIG. 18 shows a shell program of a management terminal device in the management device of the parallel computer system according to the sixth embodiment.

[Explanation of symbols]

100a to 100c: node, 101: communication processing device,
102a to 102c: communication interface mechanism, 103
a to 103c: communication cable, 104: management terminal device,
105a to 105d: System control mechanism, 106a to 1
06d: Communication cable, 107: Network concentrator, 108a to 108d: LAN control mechanism, 109a to
109d communication cable, 110 network concentrator, 200 main power supply, 201 auxiliary power supply, 202 main processor, 203 software, 204 main memory, 205 processor memory control mechanism, 206
System bus, 207: system disk, 208: I
/ O control mechanism, 209 ... RS-232C control mechanism, 21
0: bootstrap ROM, 211: system support mechanism, 212: subprocessor, 213: ROM, 2
14 SRAM, 215 local bus, 216 power on / off signal, 217 processor reset signal, 2
18 LAN controller, 219 RS-232C controller,
220: Processor, 221: ROM, 222: RA
M, 223: data interface, 224: control interface, 300: processor, 301: software, 302: main memory, 303: bootstrap ROM, 304: processor memory control mechanism, 305:
System bus, 306 I / O control mechanism, 307 System disk, 308 and 309 RS-232C control mechanism, 310 Graphics control mechanism, 311 LA
N control unit, 312 ... RS-232C control unit, 313 ... processor, 314 ... ROM, 315 ... RAM, 316 ...
Control interface, 317 ... data interface,
401 ... Adapter control command and its response, 4
02: system control command and its response, 40
3: Node message, 501: Type field, 50
2 ... source address field, 503 ... destination address field, 504 ... information field, 505 ... identifier, 601 ... type field, 602 ... source address field, 603 ... destination address field, 6
04: information field, 605: identifier, 701: packet mode, 702: non-packet mode, 703:
"SET-MODE" command, 800: disconnect state, 801: wait connect state, 802: connect state, 803: "SET-CONNECT" command, 804: call control between system control mechanisms, 1400 ...
Primary bootstrap path information, 1401 ... Alternate bootstrap path information, 105e ... System control mechanism, 106e ... Communication cable, 108e ... LA
N control mechanism, 109e communication cable, 111 management terminal device, 1700 auxiliary power supply, 1701 power-on logic, 1702 power control signal, 1703 main power supply, 17
04: terminal device, 1705: network.

Claims (1)

[Claims] 1. A management apparatus for a parallel computer system having a management terminal device for managing a parallel computer system connected to a plurality of nodes, wherein the plurality of nodes operate by a main power supply of the node and perform parallel processing. And a sub-processor that operates on the auxiliary power supply of the node and executes a system control command for managing the main processor, and that operates on the auxiliary power supply and communicates with the management terminal device, A system control mechanism for generating control data based on a system control command issued from the device, and a system control mechanism that is connected to the management terminal via a network different from the system control mechanism, and controls the node under the control of the management terminal. A LAN control mechanism for managing application software; An apparatus configured to communicate with a plurality of system control mechanisms of the plurality of nodes, a power control logic that operates on an auxiliary power supply, and turns on a main power supply of the management terminal device when a specific signal is input; Means for transmitting a system control command for turning on the main power supply collectively or individually to the sub-processors of the plurality of nodes when the main power is turned on by power-on logic; A control mechanism and a system control mechanism of the management terminal device are connected, and a system control command for managing a plurality of main processors of the plurality of nodes is transmitted from the management terminal device to a plurality of system control mechanisms of the plurality of nodes. The plurality of nodes that have transmitted and received the system control command execute a system control input that executes the contents of the command. Management system for the parallel computer system comprising: a face.