JPH06295289A - Booting method for plural computers - Google Patents

Abstract

PURPOSE:To boot plural nodes at a high speed by using the operating system on the memory of a completely booted computer node to boot plural nodes in parallel. CONSTITUTION:A computer node 2 starts a boot transmission-side program after loading an operating system OS 10 stored in an input/output device 8 to a main storage device 6 by a bootstrap 9a. The boot transmission-side program started by the node 2 sends the OS on the main storage to a network, and a bootstrap 9b of a node 3 receives it and successively stores it on the main storage of the node 3, thus booting the node 3. After the node 3 is booted, the bootstrap of a node 4 communicates with the boot transmission-side program of the node 2 and the node 4 is booted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a boot method for loading an operating system onto a memory in a multiple computer system having the same architecture connected to a network or a massively parallel computer system having a hardware configuration similar to the system. .

[0002]

2. Description of the Related Art As a method for loading an operating system through a line network, for example, Japanese Patent Laid-Open No. 3-1894.
There is a method described in JP-A-8. In this method, in a plurality of computers each having an input / output device, the operating system of the node having the input / output device storing the operating system to be used is connected to the computers of other nodes via the network. It is stored in the output device. That is, this method is a method in which the operating system is installed in one node and is transferred semi-automatically to other nodes in the network.

Further, as a method of accelerating the IPL, for example, when an external storage device that can be shared by a plurality of computer systems is provided, a node whose boot has been completed transfers the operating system to the external storage device, and another The node is intended to speed up the IPL by transferring the operating system on the external storage device to the main storage device of the node (see Japanese Patent Laid-Open No. 1-150963).

[0004]

In the former technique described above, booting is performed from the local I / O device of each node, and a node having no I / O device boots the operating system from the I / O device of another node. Even if you are using the operating system on the I / O device.

By the way, reading data from an input / output device generally requires ten to one hundred times as long as reading data from a memory. Also, in a massively parallel computer system in which thousands or tens of thousands of nodes are connected, the reliability of input / output devices is low and the failure rate is high, and it is difficult to secure a place to install the input / output devices. Since most of the nodes do not need I / O devices to perform numerical calculations, I / O devices are installed only on some nodes and local I / O devices are installed on many other nodes. There is no configuration.

From the above, when the above-mentioned conventional technique is applied to the massively parallel computer system, many nodes boot from the input / output devices of some of the nodes with the input / output devices. The performance becomes a bottleneck, and it takes a huge amount of time to finish booting the entire massively parallel computer system.

Further, the latter technique described above requires an external storage device that can be shared by a plurality of nodes, and cannot be applied to a configuration in which workstations are connected by a LAN. Furthermore, the IPL described above can be added to a massively parallel computer system in which thousands to tens of thousands of nodes are connected by a high speed network.
When the method is applied, the access to the shared external storage device is congested and the speedup is hindered.

It is an object of the present invention to provide a method for enabling a high speed boot by booting a plurality of nodes in parallel by using an operating system on the memory of a computer node whose boot has been completed.

[0009]

In order to achieve the above object, in the invention according to claim 1, a plurality of computer nodes having the same architecture are connected to a network, and some of the plurality of computer nodes are connected. The node is provided with an input / output device, and in the method of loading the same operating system on the memory of all the computer nodes, the first computer node having the input / output device is the operating system stored in the input / output device. The system is booted, the boot transmission side program is activated by the activated operating system, and the boot transmission side program sends the operating system on the memory of the first computer node to the network, The second computer node which does not have the boot reception side program starts up, The second computer node, which has completed the boot, communicates with the boot transmitting side program of the first computer node, and the boot receiving side program loads the operating system received from the network onto the memory of the second computer node. Executes the boot transmission side program, transmits the operating system on the memory of the own node to the non-booted third computer node, loads it on the memory of the third computer node, and finishes booting the computer. The feature is that the same number of computer nodes as the number of nodes are booted in parallel.

According to a second aspect of the present invention, the boot transmission side program decomposes the text and data of the operating system in the memory of the computer node into packets with respective identifiers and sequentially sends them to the network. , The size of a data area having no initial value of the operating system is stored in a packet with an identifier and sent to the network, and the boot receiving side program receives the received text according to the identifier of the packet received from the network. And storing the data in a specified position in the memory and securing the size of the data area in the memory.

[0011]

The node having the input / output device first boots the operating system stored in the input / output device, and then starts the boot transmission side program. This boot transmission side program decomposes the text and data of the operating system in the memory of its own node into packets with respective identifiers, sends them out sequentially to the network, and a data area (bss) that has no initial value of the operating system. Is stored in a packet with an identifier attached to its size and sent to the network. On the other hand, other nodes have a small-sized boot reception side program that can be resident in the memory in the ROM memory and execute this program. The boot reception side program stores the received text and data in the designated position of its own memory according to the network connection and the identifier of the packet received from the network, and prepares an area of that size for its bss in its own memory. The boot receiving side program and the boot transmitting side program cooperate with each other to realize a high-speed boot, which allows the performance of the input / output device and the number of the input / output devices to be limited, and many Node boots fast.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be specifically described below with reference to the drawings. FIG. 1 is a system configuration diagram of the present embodiment. In the figure, a plurality of computer nodes 2, 3, 4, 5 are connected to a network 1. Computer node 3,
4, 5 and 5 have the same configuration, and the CPU 11 and the main storage device 6
And bootstrap ROM 9b and LAN controller 1
It consists of 2.

The computer node 2 is provided with an input / output device 8 via an I / O adapter 7 in addition to a CPU 11, a main storage device 6, a bootstrap ROM 9a and a LAN controller 12. Then, the operating system 10 stored in the input / output device 8 is loaded into the main storage device 6 by the bootstrap 9a on the ROM, and the operating system is activated. Further, in the computer node 2, the boot transmission side program is activated from the operating system 10.

In the computer node 3, when the power is turned on, the bootstrap 9b (boot receiving side program) on the ROM is activated, communicates with the boot transmitting side program of the node computer 2, and the main storage device 6 of the computer node 2 is started. The above operating system 10 is the computer node 3
In the main memory 6 of.

Similarly, the computer node 4 is the computer node 2
To load the operating system 10, and the computer node 5 communicates with the computer node 3 to load the operating system 10. in this way,
In this embodiment, the boot processing of the computer node 4 and the computer node 5 can be executed in parallel. Since as many computer nodes as the number of computer nodes that have completed booting can be newly booted in parallel in one boot time,
The entire massively parallel system can be booted extremely fast.

Further, in the present embodiment, the operating system loaded in the main storage device is used as the transferred program, so that the operating system can be transferred without accessing the input / output device. Therefore, since it is not necessary to access the input / output device, it is possible to send the operating system to the network at a speed which is ten to 100 times as fast as when accessing the input / output device. In addition, since a computer node having no input / output device can be a transfer source of the operating system, it is possible to boot computer nodes in the order of exponentiation of boot time.

The operation of sending the operating system loaded on the main storage device as the transferred program will be described in detail below. FIG. 2 shows the structure of the operating system on the input / output device 8. The operating system is stored in the input / output device 8 in the format of management information 21, OS text 22, and OS data 23. The bss size in the management information 21 indicates the size of the variable area that has no initial value used by the operating system.

When this operating system is booted, the operating system is expanded on the main memory of the computer in the format shown in FIG. In other words, the OS text 31 and the OS data 32 are designated by the boot receiving side program at positions according to their respective hardware architectures, and are loaded at these positions. As the bss area, the boot receiving side program prepares an area for the bss size in the management information 21 as the bss 33. After that, the boot receiving program passes control to the start position of the operating system, and the OS text creates the page table 34 and the stack 35 to complete the initialization.

FIG. 4 is a processing flowchart of the boot transmission side program. First, in step 101, a communication connection is established to send out the operating system. Then, in step 102, the OS text portion 31 is sequentially stored in the data portion 42 of the packet in the format shown in FIG. In the identifier 41, an identifier indicating that the OS text 31 is stored in the data part 42 is set. The OS text portion 31 is thus sent to the network as a plurality of packets of data.

Next, at step 103, the OS data section 32 is similarly sent to the network 1 as a plurality of packets. In the identifier 41, an identifier indicating that the OS data 32 is stored in the data section 42 is set. In step 104, the bss 33 stores only the size in the data portion 42 of the packet, and the identifier 41 stores the data portion 42.
An identifier indicating that the size of bss is stored in is set. Since the size of bss is managed by the operating system, it can be obtained by the boot receiving program inquiring of the operating system. Then, in step 105, the established connection is disconnected, and the process ends.

FIG. 6 is a processing flowchart of the boot receiving side program. In step 201, first, wait for a connection establishment request from the boot transmission side program,
When the establishment request is received in step 202, the connection is established. Then, in step 203, the transmitted packet is received. The format of this packet is the format shown in FIG.

In steps 204, 206 and 208, the packet identifier 41 is determined to determine whether the received data is OS text, OS data or bss size. In step 205, the received data is O
If it is the S text, the data is stored in the OS text storage position on the main memory shown in FIG. In this embodiment,
The OS text is stored from the address "000000" on the main memory. Since the OS text is sent in a plurality of packets, it is sequentially stored starting from the address "000000" in the main memory.

In step 207, if the received data is OS data, that data is stored in the OS data storage location on the main memory shown in FIG. In this embodiment, OS data is stored from the address "400000" on the main memory. OS
Since data is sent in a plurality of packets, it is sequentially stored starting from the address "400000" on the main memory. If the received data is the bss size in step 209, an area of the bss size length in the data is secured from the next address at the end of the OS data storage position on the main memory,
Clear the area. Since the packet storing the bss size is the last packet to be received, step 2
At 10, the communication connection is disconnected. Then, in step 211, the control is passed to the start position of the OS text already stored in the main memory. The OS text prepares the necessary page table and stack on the main memory.

As a result, the operating system is loaded on the main memory of the receiving node in the format shown in FIG. 3, and booting in the computer node having no input / output device is completed.

The present embodiment is not limited to the above-mentioned one. For example, a node whose boot has been completed transmits the operating system in the memory of its own node to the entire network by using the broadcast function, and The boot may be executed at high speed.

[0026]

As described above, according to the invention described in claim 1, when a plurality of computer nodes having the same architecture are connected to the network and all the computer nodes are booted, the computer for which the boot is completed is completed. Since all computer nodes are booted by sequentially copying the operating system on the memory of the node to the memory of another computer node, the log of the total number of computer nodes N
The whole can be booted quickly with a boot time on the order of N (however, the bottom is 2).

According to the second aspect of the invention, the OS text, the OS data, and the size of the variable area having no initial value are transferred by the cooperation of the boot receiving side program and the boot transmitting side program, and the size of the variable area on the main storage device is transferred. Since it is stored in a predetermined position, high-speed boot is realized.

[Brief description of drawings]

FIG. 1 is a system configuration diagram of the present embodiment.

FIG. 2 shows a structure of an operating system in the input / output device.

FIG. 3 shows a format of an operating system in which a boot is completed on a main storage device.

FIG. 4 is a processing flowchart of a boot transmission side program.

FIG. 5 shows a packet format for operating system communication.

FIG. 6 is a processing flowchart of a boot receiving side program.

[Explanation of symbols]

 1 Network 2, 3, 4, 5 Computer Node 6 Main Storage Device 7 I / O Adapter 8 Input / Output Device 9a, 9b Bootstrap ROM 10 Operating System 11 CPU 12 LAN Controller

Claims (2)

[Claims] 1. A plurality of computer nodes of the same architecture are connected to a network, some of the plurality of computer nodes are provided with an input / output device, and the same computer is installed in the memory of all the computer nodes. In the method of loading an operating system, a first computer node having an input / output device boots an operating system stored in the input / output device, activates a boot sender program by the activated operating system, The boot transmitting side program sends the operating system on the memory of the first computer node to the network, and the second computer node having no input / output device activates the boot receiving side program, Communicates with the boot sender program of the computer node to receive the boot The receiving side program loads the operating system received from the network onto the memory of the second computer node, and the second computer node, which has completed booting, executes the boot transmitting side program, and executes the operating system on the memory of its own node. Characterized in that the system is transmitted to a non-booted third computer node, loaded into the memory of the third computer node, and the same number of computer nodes as the boot completed computer nodes are booted in parallel. Booting method for multiple computers. 2. The boot transmission side program decomposes the text and data of the operating system in the memory of the computer node into packets with respective identifiers, and sequentially sends them to the network, and the initial value of the operating system is set. The size of the data area having no data is stored in a packet with an identifier and sent to the network, and the boot receiving side program specifies the received text and data in the memory according to the identifier of the packet received from the network. 2. The boot method for a plurality of computers according to claim 1, wherein the boot area is stored in a predetermined position and the size of the data area is secured in the memory.