JP2913767B2 - Complex computer system and control method thereof - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a multi-computer system and a control method for smoothly moving a high-order process in the multi-computer system between computers.

[Conventional technology]

A multi-purpose computer system means that general-purpose large computers, small and medium computers, workstations, personal computers, etc. are interconnected by a communication device such as a local area network (hereinafter referred to as LAN) and work together with each other. This is a one-component system that can be used.

2. Description of the Related Art Conventionally, in a complex computer system, there have been various problems for effectively performing a joint work by a plurality of computers. As one of them, there is a problem that a process is smoothly moved between computers.

Generally, it is necessary to move a process between computers (this is called process migration) in the following cases.

(B) When the computer on which the original process was performed or the process itself has malfunctioned due to a program failure or hardware failure, so that another process takes over the processing of the original process on the new computer. It is. In this case, in order to improve the availability of the multifunction computer system, a method of grouping processes or jobs and backing up each other is widely used. (B) In a multi-computer system, a process moves by itself to improve the processing capacity by averaging the load or to receive a service that can be performed only by a special computer. For example,
Processes running on workstations are high
In order to obtain CPU power, the user may move to a supercomputer in the same multi-computer system.

In such a case, the process is moved. In addition, as a literature about the movement of the process, for example,
AS Tanenbaum et al., Distributed Operating System
s)] ACM Computing surveys December 1985 issue.

Further, as a conventional proposal relating to this, for example, Japanese Patent Application Laid-Open No.
There is. According to the technique described in the above publication, in a set of computers in which a plurality of processors are connected to one node, a control processor, which is one for each node, is provided with a line switching mechanism and a coupling mechanism between processors (equivalent to a LAN). , The other node or terminal behaves as if the node is a single processor system.

In addition, there is Hitachi's software product "DB / DC high reliability function ARF (Advanced Reliability Feature)". This is described in the manual “Program Product VOS3 DB / DC High Reliability Function ARF” 8090-6-334 issued by Hitachi, Ltd.

In the above software products, in the online DB / DC (database / data communication) system, the backup job is put on standby (hot standby) for each job, and the backup job is immediately taken over in the event of a failure. Increases availability as a computer system.

[Problems to be solved by the invention]

However, the following problems also exist in the devices and products of the above-mentioned documents.

First, in the computer aggregate coupling apparatus described in Japanese Patent Application Laid-Open No. 1-130252, (a) it is premised that it is realized in a connection-oriented SNA (Systems Network Architecture). The advantage of a communication device such as a LAN that can perform information communication cannot be used.

(B) Special and complicated hardware and software such as a line switching mechanism and a coupling mechanism are required.

(C) A single control processor distributes communication messages to each processor in the computer aggregate, manages location information of resources in the computer aggregate (directory service), and states of each processor (live or dead)
Monitoring and the like are all executed independently, the performance and availability of the control processor limit the performance and availability of the entire computer aggregate.

Next, the method of taking over the process performed by the DB / DC high reliability function ARF (Advanced Reliability Feature) is as follows: (a) The new process takes over the database, journal (log), etc. from the failed process Continue online processing. However, the lines and terminals used by the old process cannot be used as they are, and it is necessary to newly perform connection processing with them. As a result, from the point of view of the user in front of the terminal, when the process is switched, it is felt that the connection with the computer is once disconnected and then the connection is restored, which is desirable in that the process is easily moved. Absent.

In a large online system, the number of connected terminals may be in the thousands or tens of thousands, so it takes a considerable amount of time for a newly started process to move the process to restore connection with all of those terminals. Will take.

(B) Another problem associated with the movement of a process is that a partner terminal or another process communicating with the old process knows the fact that the target process has moved and the network address of the computer to which the process has moved. There is a need.
Normally, message transmission to a process in a computer system is rarely performed directly by specifying the physical network address of the computer on which the target process is located. Generally, directory servers, name servers,
Alternatively, a device called a gateway or the like may be configured to include the name of the target process (identifier used by humans) or a logical network address (for example, a DARPA internet address) assigned independently of the physical properties of the transmission medium. Manages the network address set of the computer where the target process is located.

Here, an example of a process migration in a compound computer system managed by a name server will be described. The process started on the new computer first informs the name server that its location has changed. Upon receiving this, the name server rewrites the set of the process name and the network address for the process, and if necessary, notifies the other name servers holding the same information as well. Thus, after that,
The terminal that has obtained the network address of the process from the name server to communicate with the process will obtain the network address of the new computer,
It is possible to send to the correct destination. However, the following problems also occur in this system.

(A) There is a slight time lag (delay) from when the process moves to when the information of the name server relating to the process is actually updated. This is particularly noticeable when the same information is duplicated and held by a plurality of name servers, or when the size of the network is large. As a result, transmission may be performed to a network address for which a target process does not already exist, according to the old information of the name server.

(B) Some computer networks do not refer to a name server. That is, in a terminal or the like as simple hardware, transmission and reception are performed using only a physical network address, and thus the above-described mechanism does not function.

(C) Centralized management of network addresses by the name server inevitably leads to concentration of access to the name server, which limits the capacity and availability of the entire multifunction peripheral system.

As described above, in the conventional multifunction computer system, it is extremely difficult for the process that has moved the computer to continue the communication with another process, the terminal, or the like that was communicating before the move.

An object of the present invention is to solve such a conventional problem, and a process in which a computer is moved smoothly moves to a new computer without affecting other processes and the like which were communicating with the process immediately before. And a control method therefor.

[Means for solving the problem]

In order to achieve the above object, a method of controlling a multi-functional computer according to the present invention is characterized in that (a) a single network address that is identifiable from another interface device is allocated to one interface device, and (a) the network address is allocated. Single station selection mode, which extracts only communication messages addressed to the network address from the transmission medium and passes them to the upper process.
(B) a group selection mode in which a group of a plurality of network addresses is defined as one group, and only communication messages addressed to the group are extracted from a transmission medium and passed to a higher-level process; and (c) all communication flowing on the transmission medium. A non-selection mode in which a message is extracted regardless of a destination and passed to a higher-level process is provided, in whole or only in part. It is characterized in that it is arbitrarily changed by instructing the interface device. Further, (b) the interface device is characterized in that, while operating in the group selection mode or the non-selection mode, the higher-level process is notified of the network address to which the received communication message is originally directed. is there. (C) When transmitting a communication message, the interface device (d) uses a network address originally assigned to the interface device as a transmission source network address of the communication message, The host process includes only one or both of modes using a value assigned as a source network address from a process, and the higher-level process stores the interface in the operation mode and the source network address in the case of (e) above when the network is operating. There is also a feature in that it can be arbitrarily changed by instructing the apparatus. Further, (d) the higher-level process accesses the interface device via the operating system, and the access uses the single station selection mode of (a) and the network address originally assigned to the interface device of (d). It is characterized in that it is executed in the same way as access to an interface device that has only a mode to be used and does not have a function of changing the target network address. further,
(E) The upper process is also characterized in that it has a memory for storing information about other upper processes and terminals in communication, and a communication means for transferring the information stored in the memory to another upper process.

[Action]

In the present invention, the process of moving a computer is realized by a destination computer receiving a communication message addressed to the original computer. Therefore, the interface device to the transmission medium of the network arbitrarily changes the range of the communication message to be received according to the instruction of the upper process when the network operates. As a result, the process that has moved the computer can use the same network address as that of the original process in the destination computer, and other processes that have communicated with the original process using the network address can be used. And terminals,
It is not affected by the movement of the process.

〔Example〕

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 and FIG. 2 are block diagrams of a multifunction computer system showing one embodiment of the present invention.

This multi-functional computer system operates as an online DB / DC (database / data communication) system, and includes computers 13, 14, 15 and an interface device 1
A LAN is composed of 1, 12 and the transmission medium 16. That is, 1
1, 12 are LAN interface devices, 13, 14, 15 are upper processes in a computer, 16 is a LAN transmission medium, 17 is a wide area network interface device, 18 is a wide area network (for example, a public telephone communication network is included). , 19 are terminals accommodated in the wide area network 18.

For example, Ethernet (Xerox LAN Eth
In the case of ernet (trademark), the LAN transmission medium 16 is a coaxial cable, the LAN interface devices 11 and 12 are transceivers and a controller board, and the network address is 6 bytes defined by a MAC (medium access control) sublayer protocol. , Respectively.

FIG. 2 shows a takeover procedure when the process 14 backs up the process 15 due to a failure or the like during communication between the process 15 and the remote terminal 19, and FIG. 1 shows the procedure after taking over according to the procedure of FIG. This indicates the movement status of the process.

In FIG. 2, the upper process 15 (online DB / D
C process) is connected to a remote terminal 19 via the wide area network 18 to perform online operations.
The upper process 14 is waiting for backup of the upper process 15. In the LAN interface device 11, a batch job 13 that does not require relatively real-time processing is connected and executed in addition to the host process 14. In this case, the LAN interface device 11 extracts only a communication message addressed to a predetermined single network address from the network transmission medium 16 and
It is operating in the mode passed to 14, 14 (hereinafter referred to as single station selection mode).

In FIG. 2, the upper process 15 is a remote terminal 19.
And establish a connection (step 1), and thereafter start an online business. At the same time, the upper process 15 transmits information necessary for taking over the business to the backup process 14 in case of a failure (step 2). The information necessary for the handover includes information relating to connection with each terminal (session context), a database update history (journal) generated in online business, information on lock to the database, and the like. Also, the number of times that the takeover information is transmitted may be only once at the start of the online business, or may be periodically transmitted a plurality of times during the business. Further, as a method of giving the takeover information to the backup process 14, as shown in FIG. 2, in addition to a method of transmitting the data via a LAN, for example, a method of transmitting the data via a shared magnetic disk or the like is also available. is there. That is, it can be realized by writing the takeover information from the upper process 15 to the magnetic disk and periodically reading it out by the backup process 14. When, how, and what is given as the takeover information from the higher-level process executing the business to the backup process largely depends on the realization policy and the realization method of the fault tolerance function in the multifunction peripheral system.

In FIG. 2, it is assumed that the terminal 19 performs a session (connection, that is, a procedure for "starting communication") with the online DB / DC process 15 or 14.
It is also assumed that, in order for the backup process 14 to take over the business, it is necessary to take over information negotiated (negotiated) on the session at the time of setting the session and information such as the terminal status managed by the original process 15. . if,
Terminal 19 and higher-level process 15 are connectionless type (type that immediately communicates without an agreement to "start communication")
In the case where a communication method is performed and the upper process 15 does not hold the terminal state, the above-described steps 1 and 2 are unnecessary, and the process takeover process is extremely simplified.

Next, referring to FIG. 1, the host process 15 becomes inoperable due to a failure such as a program failure (step 3). The backup process 14 detects the failure by any method, and determines that the backup process 14 takes over the online operation of the upper process 15. Backup process 14
As a method of detecting a failure of the original process 15, for example, there is a method of periodically sending a survival certificate (I am alive) message to each other. For details of this method, for example,
It is described in J. Gray et al., "Fault Tolerance System", edited by Eiichi Watanabe, published in 1986 by The McGraw-Hill Book.

The backup process 14 depends on the LAN
When transmitting / receiving a communication message, the interface device 11 is instructed to change the address used as its own network address to the address 2 used by the original process 15 (step 4). Thereby, the backup process 14 receives the communication message transmitted from the remote terminal 19 to the original process 15, and further uses the network address used by the original process 15 as the source network address. Thus, it is possible to transmit to the terminal 19. This means that the backup process 14 is completely
It is possible to perform an operation indistinguishable from. Also, since the information required for taking over the online business was transmitted to the backup process 14 in step 2,
The handover is performed smoothly, and the service to the terminal 19 is practically not stopped.

At this time, the LAN interface device 12 stops the LAN transmission medium 16 because the upper process 15 is stopped.
Do not send communication messages to. But if LA
N interface device 12 accesses the LAN transmission medium 16 independently, which is the upper process that has taken over the online business
When there is a possibility of adversely affecting the processing performed by the LAN interface device 14 and the LAN interface device 11, for example,
When the operator intervenes, an operation such as stopping the LAN interface device 12 or disconnecting from the LAN transmission medium 16 is required.

Finally, a return procedure when the upper process 15 has recovered from the failure will be described. When the upper process 15 recovers from the failure, it multicasts a communication message indicating that it has recovered on the LAN transmission medium 16. Upon receiving this communication message, the upper process 14 determines to return to the original backup mode. Note that a communication message to be mastcast to the LAN transmission medium 16 may be transmitted only to the address 2, for example. Furthermore, a method is also possible in which the upper process 14 knows the recovery of the original upper process 15 by operator intervention.

In any case, the upper process 14 instructs the LAN interface device 11 to return the network address to the original address 1. As a result, the composite computer system returns to the original operation state as shown in FIG.

In the above description, there is only one disadvantage. That is, the network address (address 1) of the LAN interface device 11 is changed to the address 2 by the upper process 14.
Therefore, in this state, reception of a communication message transmitted to the address 1 originally held by the LAN interface device 11 becomes impossible in this state. Also, when there is an important upper process other than the backup process 14 when communicating using the LAN, the process 14 may not be able to monopolize the LAN interface device 11 and take over. obtain. In such a case, an operation to be taken by the host process 14 or the LAN interface device 11 in this embodiment will be described.

FIG. 3 is an explanatory diagram of the operation of the multifunction computer system according to another embodiment of the present invention, in which a set of a plurality of network addresses is defined as one group.

The configuration of the multifunction computer system shown in FIG. 3 is the same as that shown in FIG. 1, and the backup process 14 takes over the online business because the online DB / DC process 15 has failed. . But the third
In the figure, the LAN interface device 11 can define a set of a plurality of network addresses as one group, and extracts only communication messages addressed to a certain group from the transmission medium 16 of the network, and It operates in an operation mode of passing (hereinafter, referred to as a group selection mode). Backup process 14
Specifies the address 1 and the address 2 to the LAN interface device 11 as a group, and instructs the LAN interface device 11 to receive a communication message addressed to the group (step 6).
As a result, the upper processes 13 and 14 can take over the processing of the failed process 15 and continue their original processing (step 7). That is, the communication message transmitted from the terminal 19 to the batch job 13 via the LAN interface device 11 also
Online DB / DC from LAN interface device 11
The communication message transmitted independently to the process 14 is also received and processed together with the communication message addressed to the process 15 as a backup process.

Further, in the same situation, the LAN interface device 1
1 operates in a mode in which all communication messages flowing on the transmission medium 16 of the network are extracted regardless of their destinations and passed to the upper processes 13 and 14 (hereinafter, this is referred to as a non-selection mode). It is possible. In this case, the same effect as in the previous group selection mode can be obtained. Of course, the host process 14 instructs the LAN interface device 11 to receive all communication messages.

As described above, the examples of the group selection mode and the non-selection mode have been described. However, a problem also occurs in these operation modes. That is, in any of these cases, after receiving the communication message from the LAN interface device 11, the upper processes 13, 14 need to check whether the communication message is originally intended only. Further, when transmitting, it is necessary to transmit to the LAN interface device 11 whether or not it is originally sent from (the network address of the transmission source). This imposes new processing overhead on both the upper processes 13, 14 and the LAN interface device 11. Further, as is apparent in the case of the non-selection mode, the amount of communication messages handled by the upper process 14 is larger than that in the case of the single-station selection mode.

FIG. 4 is a processing flowchart when a LAN interface device having three operation modes receives a communication message.

The LAN interface device 11 constantly monitors the LAN transmission medium 16, and when a valid communication message frame (or packet) appears on the transmission medium 16, it checks the designation of the destination network address (step 41). If it is a broadcast communication (broadcast or multicast), a communication message is passed to the upper process 14 (step 42). If it is not a broadcast communication, the operation mode of its own (LAN interface device 11) is checked, and if it is a non-selection mode, the communication message is passed to the upper process 14 regardless of the destination (step). 43). If the mode is not the non-selection mode, it is determined whether or not the mode is the group selection mode (step 44). If the mode is the group selection mode, whether or not the received communication message is addressed to a computer included in the group is determined. (Step 45),
If so, the communication message is passed to the upper process 14. If it is not the group selection mode, since it is the single station selection mode, it is checked whether or not the received communication message is addressed to itself (step 46).
Pass the communication message to. If it does not apply to all, do not pass the communication message to the upper process 14,
Discard (step 48).

When checking whether or not the LAN interface device 11 is addressed to itself, it is determined whether or not the destination network address of the received communication message matches the network address held by the LAN interface device 11, If they match, it is addressed to you.

FIG. 5 and FIG. 6 are explanatory diagrams of a control method by an operating system of a multifunction computer system according to another embodiment of the present invention.

In the embodiment shown in FIG. 1 to FIG.
Various instructions for operating the operation mode and the network address have been issued to the LAN interface device 11. However, in the present embodiment, all these instructions are performed by the operating system (hereinafter, referred to as OS), thereby reducing the load on each upper process.

The idea of the present embodiment is that the LAN interface device is a port in another name, and is one of the computer resources managed by the OS like the instruction processor and the memory, and in detail, is one of the resources to be dispatched. That's what it means.

In FIG. 5, upper processes 54 to 56 executed by three instruction processors 57 to 59 in the multifunction computer system include:
It shows a state in which it is operating uniformly by being connected to the LAN transmission medium 60 via the LAN interface devices 51 to 53. And OS61 that manages the complex computer system
Are upper processes 54 to 56 and instruction processors 57 to 59, and upper processes 54 to 56 and LAN interface devices 51 to 53.
It intervenes between and controls each operation. In this case, the OS
The LAN interface device 51 and the instruction processor 57 are assigned to the upper process 54 by 61. Similarly, LAN interfaces 52 and 53 and instruction processors 58 and 59 are also assigned to the upper processes 55 and 56, respectively. Therefore, the upper processes 54 to 56 are equally receiving the service of the computer resources. Here, the LAN interface devices 51 to 53 operate in the single-station selection mode.

FIG. 6 shows a case where it is necessary to allocate more computer resources to the upper process 54.

For this purpose, the OS 61 rewrites the network address of the LAN interface device 52 so as to be the same as the LAN interface device 51. Then, the LAN interface device 52 and the instruction processor 58 are taken from the upper process 55 and assigned to the upper process 54. As a result, part of the access to the LAN interface device 51 performed by the upper process 54 (that is, transmission / reception processing of the communication message) is performed by
61 automatically converts the access to the LAN interface device 52. As a result, the upper process 54 can transmit and receive communication messages that exceed the transmission capacity (bandwidth) of a single LAN interface device, for example.

〔The invention's effect〕

As described above, according to the present invention, in a multifunction computer system, a destination process has no effect on other terminals or processes that have been communicating with the process up to immediately before, and has no effect on other computers. It can move smoothly.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 and FIG. 2 are explanatory diagrams of process movement of a multi-functional computer system showing one embodiment of the present invention, and FIG. 3 is a multi-functional computer system showing another embodiment of the present invention. FIG. 4 is an explanatory diagram of a network address designation method, and FIG.
FIGS. 5 and 6 are flowcharts of a process performed by the interface device. 11, 12, 51 to 53: LAN interface device, 13 to 15, 54 to 56:
Upper processes, 16, 60: LAN transmission medium, 17: wide area network interface device, 18: wide area network, 19: terminal, 57 to 59: instruction processor, 61: operating system (OS).

Claims (6)

(57) [Claims] 1. A transmission medium, comprising: a transmission medium; two or more interface devices connected to the transmission medium; and a plurality of computers executing a high-order process operating as a user of the interface device. In the network, one interface device is assigned a single network address that can be distinguished from another interface device, and (a) extracts only a communication message addressed to the assigned network address from the transmission medium. Single station selection mode to pass to the upper process
(B) a group selection mode in which a group of a plurality of network addresses is defined as one group, and only communication messages addressed to the group are extracted from the transmission medium and passed to a higher-level process; A non-selection mode that extracts the communication message regardless of the destination and passes it to the higher-level process, comprising all or only a part,
A control method for a multifunction computer system, characterized in that a higher-level process instructs the interface device when the network is operating, and arbitrarily changes the set of the three operation modes, the network address, and the network address. 2. The method according to claim 1, wherein the interface device notifies an upper process of which network address the received communication message is originally intended for while operating in the group selection mode or the non-selection mode. The control method for a multi-function computer system according to claim 1. 3. The communication apparatus according to claim 1, wherein: when transmitting the communication message, the interface device uses (d) a network address originally assigned to the interface device as a transmission source network address of the communication message; And / or only a mode using a value assigned as a source network address from a higher-level process. The higher-level process sets the operation mode and the source network address in the case of (e) above when the network is operating. 2. The control method according to claim 1, wherein the instruction is changed arbitrarily by giving an instruction to the interface device. 4. The high-level process accesses the interface device via an operating system, and the access is performed by being originally assigned to the (a) single station selection mode and the (d) interface device. 2. The control method of the multi-computer system according to claim 1, wherein the method is performed in the same manner as the access to an interface device having only a mode using the network address and having no function of changing the target network address. . 5. A transmission medium, comprising: a transmission medium; two or more interface devices connected to the transmission medium; and a plurality of computers for executing a high-order process operating as a user of the interface device. In the network, the higher-level process has means for storing information on another higher-level process and a terminal in communication, and means for transferring information stored in the higher-level process to another higher-level process. Computer system. 6. A transmission medium, comprising: a transmission medium; two or more interface devices connected to the transmission medium; and a plurality of computers for executing a high-order process operating as a user of the interface device. In a network, a single network address that is identifiable from other interface devices is assigned, and (a) only a communication message addressed to the assigned network address is extracted from the transmission medium and passed to a higher-level process. A station selection mode; (b) a group selection mode in which a group of a plurality of network addresses is defined as one group, and only communication messages addressed to the group are extracted from the transmission medium and passed to a higher-level process; Extract all communication messages flowing on the medium regardless of the destination and send them to the upper process All or only a part of the non-selection mode, the set of the three operation modes, the network address and the network address from the upper process is arbitrarily changed by being instructed when the network is operating, While operating in the group selection mode or non-selection mode, the received communication message is originally intended for which network address,
When notifying the higher-level process and transmitting the communication message, (d) the interface device is originally designated as the transmission source network address of the communication message.
An interface device comprising: a mode using an assigned network address; and (e) a mode using a value assigned as a transmission source network address from a higher-level process, or only one of the modes.