JPS61267859A - Decentralized processing system - Google Patents

Abstract

PURPOSE:To avoid the useless message transmission and to improve the message exchange efficiency by reflecting the state of a process having a dynamic transi tion onto a process control table PMT and confirming the state of the process at the remote side on the PMT. CONSTITUTION:A decentralized processing system is formed with three nodes consisting of personal computers 10, 20 and 30 and network definition files 19, 29 and 39 which are connected with each other via a local area network LAN100. The superviser processes 12, 22 and 32 start the processes 11, 21 and 31 that should exist on the own node according to the process definition information written to the files 19, 29 and 39 and also perform the communication with the processes on another node in response to the requests given from the processes 11, 21 and 31. Furthermore the processes 12, 22, 32 perform the registration or up dating of PTM13, 23 and 33. The execution of these processes 12, 22, 32 are decided with reference to the PMT.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a distributed processing system, and particularly to a distributed processing system in which processes existing on a plurality of nodes perform processing through message exchange. More specifically, the present invention relates to a process management method in such a distributed processing system.

(Prior Art) The configuration of a combinator network can be roughly divided into two types, focusing on the arrangement of processing units and the division of functions. ) a distributed type, a horizontal distributed type that distributes equally among multiple sefters, and a composite distributed type that combines these types.
Become a seed. The horizontally distributed type is superior to the vertically distributed type in terms of flexibility in system operation and load uniformity, especially in recent years when information processing is becoming more oriented toward end-users (integrated circuits are becoming increasingly popular). With the combination of technology and fractional processing technology (seeds), horizontally distributed processing systems such as a plurality of personal combi-complexers or office combiators interconnected via local communication lines have been actively adopted.

In this type of distributed processing system, the physical elements that make up the computer network, such as personal computers, are treated as logical devices, or nodes, and the communication lines that interconnect each node are used as logical transmission paths, or , are modeled as links, and the world of combinator networks is expressed as a combination of nodes and links.

Further, various user programs (or application game exploit programs) and operators that attempt to communicate between nodes using the functions of these nodes and links can be regarded as logical communication entities, that is, as processes.

Processing in a distributed processing system proceeds through message exchange between processes, and at this time a new process is created or, conversely, a previously existing process disappears. Furthermore, since these processes can be the subject of message exchange for all other processes, appropriate management of the processes is important.

(Problem to be Solved by the Invention) However, in the conventional distributed processing system as described above, there is no proper management of processes that are dynamically created and processes that disappear. When sending a message, the destination of the other process and whether the message can be sent are not always certain, which can cause problems in message sending or result in unnecessary messages being sent. be.

Therefore, an object of the present invention is to accurately register management information of processes that are dynamically generated and disappear in a table, and to refer to this table when sending messages. By doing so, it is an object of the present invention to provide a distributed processing system that eliminates the above-mentioned problems and allows efficient message exchange.

(Means for Solving the Problems) The system of the present invention is a distributed processing system in which processes existing on a plurality of nodes perform processing through message exchange. Process definition information consisting of the job number, existing node address, and startup information for the process that normally exists in the processing system is stored in the network definition file written when the distributed processing system was constructed, and the status of the process on each node. , a process management table in which identification numbers, node addresses, etc. are registered, and a process definition information that activates the subekifrocess existing on the own node and communicates with processes on other nodes in response to o requests from the process. and a supervisor process that registers or updates the process management table in conjunction with startup and communication, and the supervisor process determines execution by referring to the corresponding process management table when sending a message. It is characterized by

(Example) Next, an example of the present invention will be described in detail with reference to the drawings.

FIG. 1 shows a block diagram of an embodiment of the present invention, and FIG. 2 shows a detailed block diagram of a node in this embodiment.

Referring to FIG. 1, in this embodiment, each personal combinator (hereinafter referred to as a personal computer) 10.20.3
0 and network definition file (hereinafter referred to as NDF) 1
This is a distributed processing system in which three nodes consisting of nodes 9, 29, and 39 are interconnected via a local area network (hereinafter referred to as LAN) 100.

In such a distributed processing system, each node is almost equally active.
Because it has comparable functions and configurations, it has greater flexibility in system operation and is superior in terms of load equalization compared to conventional vertical distributed processing systems.

Furthermore, it is rapidly becoming popular in offices, factories, etc. because it matches the end user orientation of recent information processing and is easy to use. Although this embodiment shows only three nodes to simplify the explanation, in reality it often consists of a larger number of nodes, for example, several nodes. Each of the personal computers 10, 20, and 30 is Each process 11, 21 and 31 based on a user program is executed. Each of these processes proceeds through message exchange via the LAN 100.
Although only one process is shown, multiple processes may exist.

In addition, each computer 10.20 and 30 levers,
It has supervisor processes 12, 22, and 32, and process management tables (hereinafter referred to as PMT) 13, 23, and 33. Each of the PMTs 13, 23 and 33 is used to register the status (available for service, in service, and unavailable for service), identification number, node address, etc. of the processes 11, 21 and 31 in the system.

Each of the supervisor processes 12, 22 and 32 starts processes 11, 21 and 31 based on process definition information on NDFs 19, 29 and 39 as described later, and also starts processes 11, 21 and 31 based on the process definition information on NDFs 19, 29 and 39, and requests from processes 11, 21 and 31. In addition to communicating with processes on other nodes in response to
Register or update MT13, 23 and 33.

Each of the NDFs 19, 29, and 39 has the same content, storing process definition information consisting of identification numbers, existing node addresses, and activation information for processes that normally exist in this distributed processing system.

This storage is performed when this distributed processing system is constructed.

FIG. 2 shows supervisor processes 12.22 and 32.
is represented by the reference numeral 2, and similarly represents the process 1.
It is a block diagram showing the connection relationship with PMT3 and NDF'9, and the detail.

Referring to FIG. 2, the supervisor process 2 has initialization processing means 2A, transmission means 2B, reception means 2C, and initialization processing means 2D.

When the supervisor process is activated, the two initialization processing means read the process definition information stored in the NDF9, register the identification number and node address of the information in the PMT3 via the table processing means 2D, Next, based on the startup information (consisting of a process name and startup parameters), the process 1 that should exist on the own node is started, and the success or failure of this startup is added to the PMT 3 via the table processing means 2B.

When the process 1 started as described above attempts to send a message to a process on another node, the sending means 2B refers to the PMT 3 via the table processing means 2D to confirm the state of the other process, If the connection described below has already been set between the other party's process and process 1, the message will be sent; if no connection has been set and the other process is serviceable, the message will be sent to the other party's supervisor process. The table processing means 2D instructs the connection setting and updates the information on the PMT 3.

Further, the receiving means 2C accepts messages sent from processes of other nodes, and the table processing means 2D
The state of the process 1 of the own node is confirmed on the PMT 3 via the PMT 3, and if a connection has already been established between the partner process and the process 1, the message is processed as process definition information. Furthermore, if the message instructs the setting of a connection between both processes and the process 1 is serviceable, it notifies the supervisor process of the other party of the connection setting and sends it to the table processing means 2D. Update the information above.

The status of process 1 was checked in the source process when sending the message, and the message was sent because it was possible to send the message in reverse.However, the status of process 1 is also checked when the message is received at the destination. Since it cannot be guaranteed that there will be no change in the process 1, it is necessary to check the status of process 1 even when receiving a message, as described above.

As described above, the table processing means 2D registers with the PMT 3 in response to each request from the two initialization processing means at the time of initialization, and from the transmitting means 2B or the receiving means 2C at the time of message transmission/reception.

View or update. Furthermore, when a process is generated or disappears while processing progresses through inter-process communication, the PMT 3 is updated in response to a request from the transmitting means 2B or the receiving means 2C.

FIG. 3 shows a processing flow performed when the supervisor 2 is started. The operation of this embodiment when the supervisor 2 is activated will be described below with reference to FIG.

First, in step (2), the supervisor process 2 fetches the own node address physically set in the node by a switch or the like from the operating system (not shown) and opens the NDF 9. Next, process definition information for one process is read from NDF9 (step ■), and if this does not signal the end of NDF9 (step ■), the identification number, node address, and status (of the read process definition information) are read. For the time being, service is enabled) as described above, and register it with PMT3 (
Step ■).

Check whether this node address matches the own node address read in step ■ (step ■), and if it matches, the process definition information read in step ■ corresponds to the process that should exist on the own node. Therefore, the process 1 is started using the start information consisting of the process name and start parameters of the process definition information (step 2).

Then, if the startup is successful (step ■), 0-1
Since the status of process 1 was determined to be serviceable in step (2), it is left as is, and a startup message is sent to all other nodes to inform that process 1 on its own node has become serviceable (step (2)). Also, step ■
When you see that process 1 has failed to start,
Set the state of process 1 to unserviceable (step [phase]).

From step ■ or step [phase], move to step ■fi, and for PMT3, step ■ or step [phase]
The state determined by phase] and the communication channel number and process number recognized by the operating system are
Add to MT3. Communication channel 11 is a channel number for supervisor process 2 to send messages to process 1, and the process number is assigned by the operating system.

From step ◎, as in the case where the address of the own node does not match in step
The process is repeated until an end op file (EOF) is detected at .

As a result of the above processing, in the PMT 3 on the node where the supervisor process 2 has been started, the job number, node address, status (
Service enabled or service disabled) 1 communication channel number and process number are registered, and for processes on other nodes, identification numbers, node addresses and status (
service) will be registered. Note that the PMT on the node that received the startup message in step
In step 3, if the process that sent the main activation message was in a non-serviceable state, it is updated to a serviceable state. Therefore, when a process that has been in a service-unavailable state due to a processing failure for some reason is restarted, it is accurately reflected in the PMT 3 on one other node.

Now, when supervisor process 2 and process 1 are activated, process 1 becomes capable of processing. This process includes not only processes that remain within the local node, but also back-communications (messages) between processes on different nodes, such as when transferring files between nodes or referring to control information. (exchange). A supervisor process attempting to send a message refers to the PMT3 on its own node, and executes the message sending only when the other process is in a communicable state.

For this purpose, it is necessary that the process status is accurately reflected in the PMT3.

As is well known, inter-process communication is carried out through various protocol commands exchanged between supervisor processes, but the one that affects the state of processes is the activity start (which instructs the start of a connection). A
activity 8 tart), an activity discard (Activity 8 tart) that instructs the abandonment of a bond.
Discard), an activity discard attack (ActivityDiscard) that notifies the abandonment
ard Ack), Activity End (Activity End) which instructs the end of the binding, Activity End Ack (Activity End) which notifies the end of the binding.
vityEnd Ack), exception data that notifies the exception event.
) and control data.
ed Data).

To start the join, the supervisor process (hereinafter referred to as supervisor A) of process all, which receives a join request from a process (hereinafter referred to as process a) by specifying the other process (hereinafter referred to as process b-j), b II supervisor process (hereinafter referred to as supervisor B)
An activity start is sent to the PMT on the supervisor A side (hereinafter referred to as PMT A), and each state of process a and process b on the PMT on the supervisor A side (hereinafter referred to as PMT A) is set to in-service.

When Supervisor B receives the activity startra, it transmits the PMT (hereinafter referred to as PMTB) on the Supervisor B side.
Check the status of process b above, and if the service is available, return type data as an affirmative response and P
A connection is set with process a and process b on MT B in service status. Further, if the state of process b on PMT B is in service or unavailable, a reason code is attached to the exception data as a negative response and a reply is sent.

When supervisor A receives type data, it sets up a connection, and when it receives exception data, it changes the state of process b on PMT A to unserviceable if the internal code is unserviceable. Eventually, in response to the activity start from the supervisor, supervisor B replies with typed data, and when supervisor λ receives this, process a and process b are combined and messages can be exchanged between them. be.

If you want to abandon the binding established in this way midway through message exchange, the supervisor process on the subject side (in this case, either Supervisor A or Supervisor B
) sends an activity discard to the other party's supervisor process with a reason code attached. In this case, in response to sending and receiving the activity discard, the supervisor and supervisor B change the state of each process a and process b on PMT A and PMT B to a state corresponding to the reason code. .

Furthermore, when the connection is terminated as the message exchange via the connection set as K is completed as described above, the supervisor process on the subject side (supervisor A in this example) transfers the activity end to the supervisor on the other side. (Supervisor B in this example).

In response to receiving the activity end, supervisor B releases the connection with the supervisor, returns the activity end book, and sets the status of process a and process b on PMTB to serviceable. On the other hand, supervisor A, which has received the activity endbook, also releases the connection with supervisor B and sets the status of process a and process b on PMT A to serviceable.

FIG. 4 shows the state transition of the process that responds to the protocol command described above, together with the response to the process startup process.

(Effects of the Invention) According to the present invention, by employing the above configuration, the dynamically changing process state is accurately reflected on the PMT. This allows message transmission to be executed only when the process ID of the partner process is IN-identified on the PMT and communication is possible, and unnecessary message transmission can be avoided and messages can be exchanged efficiently.

[Brief explanation of the drawing]

FIG. 1 shows an embodiment of the present invention, FIG. 2 shows a detailed view of the embodiment, and FIGS. 3 and 4 are diagrams for explaining the operation of the embodiment. 1.11,21.31...Process, 2,12
,22゜32... Supervisor process, 3,
13,23.33...Process management table (
PMT), 9, 19, 29゜39... Network definition file (NDF), 10゜20.30...
...Personal Combiator (PC). 2 people: Initialization processing means, 2B: Transmitting means, 2C: Receiving means% 2D:
・Table processing means, 100 Fig. 1 Fig. 3

Claims (1)

[Scope of Claims] A distributed processing system in which processes existing on a plurality of nodes perform processing through message exchange, wherein each node has a process that normally exists in the distributed processing system. A network definition file in which process definition information consisting of an identification number, an existing node address, and startup information for the process was written at the time of constructing the distributed processing system, and a process on each node, its status, the identification number, and the node. A process management table in which addresses, etc. are registered, and a process that starts a process that should exist on its own node based on the process definition information and communicates with a process on another node in response to a request from the process. and a supervisor process that registers or updates the process management table in conjunction with the activation and communication, and the supervisor process determines execution by referring to the corresponding process management table when sending a message. A distributed processing system characterized by: