JPH0619763B2 - Access method between nodes in computer system - Google Patents

Description

The present invention relates to an access method between nodes in a computer system in which two or more computers are connected by a high-speed communication line.

"Prior art" A large computer is used for the system using a computer, and a so-called centralized management system in which a large number of terminals are connected to this and independent computers are installed in various places, and these are installed in each department. There is a so-called distributed management system that is used independently. The distributed management system has many advantages in terms of interactive function, cost, and efficiency. With the low price and widespread use of personal-use computers, there is a marked increase in the adoption of such systems.

By connecting the computers that are distributed and managed by a communication line, so-called resources can be shared between the computers, and the advantages of the centralized management system can also be provided. Hereinafter, such a system will be referred to as a computer network system for the description.

In this computer network system, each computer connected to a communication line is called a node. In communication between these nodes, various communication systems have been developed to prevent signal collision on the communication line.

For example, an identification code called a token for identifying a node having a transmission right is inserted in a frame of a signal propagating in a ring-shaped communication line. Thus, only the node having the transmission right can send a signal to this communication line, thereby preventing the signal collision. By this method, data can be exchanged between nodes at a relatively high speed.

In such a system, when a node desires to use a data file or program file stored in the storage device of another node, it requests the other node to transfer the program file, etc. It can be stored again in the storage device of the node and the processing using it can be executed thereafter. As a result, it becomes unnecessary to always store all the necessary files in the storage device of its own node, and so-called resource sharing is performed.

In such a computer network system, file transfer work is executed by the following procedure, for example.

First, the first node specifies a name (node name, file name) that identifies a program file or a data file, and sends the transfer request to the communication line. The central processing unit of the second node holding the file issues a data transfer command to the device storing the file.

Specifically, the central processing unit performs an operation of checking the address of the target file included in the device, reading the address, and sequentially transferring the read address to the first node through the communication line.

“Problems to be Solved by the Invention” In the conventional computer network system as described above, even though the resource is shared, the use of the file stored in the device of another node does not mean that the file can be directly accessed. Once, the central processing unit of the node was requested to transfer the file.

Therefore, during this data transfer process, the central processing unit of the node on the request side is monopolized, and the task of the node itself is hindered.

Of course, there is also a so-called DMA (Direct Memory Access) method in which the central processing unit performs only the search of the head address of the file and then the device automatically performs the data transfer operation.

However, in any case, when data transfer work is performed in file units, line occupation time increases and waiting time increases,
There was a drawback that the dialogue processing mechanism was damaged.

Also, on the side requesting the transfer of the file, it is necessary to secure a memory having a sufficient capacity for temporarily storing the file. However, when working with a large program or a large amount of data, the memory capacity is often insufficient, which makes the work impossible. Become.

Files that are read and written in this way are
It is not always in a so-called disk memory for data storage. The file exists also in the main memory, the image memory of the display unit, or the like, and writing and reading can be performed in the same manner. In the present invention, devices having such files will be generically referred to as resources hereinafter. That is, some resources have only one file and some devices have multiple files.

On the other hand, in a computer network system, for example, when there is a program that performs a complicated operation using a large number of files, it is necessary for a node that holds the file and has special hardware suitable for executing the program. It is also performed to transfer instructions, request arithmetic processing, and receive execution results.

This is conventionally known as a so-called remote job entry system and is utilized, for example, for searching a database. However, it has not been easy for a central processing unit of an arbitrary node to incorporate such an execution request processing into a part of a process to be processed at any time and control the execution place thereof.

The present invention has been made to solve the drawbacks of the conventional computer network system as described above, and a node connected to a communication line can transfer files in the resources of other nodes to the same extent as those of its own node. An access method between nodes in a computer system that can process at high speed is provided.

"Outline of the Invention" A computer network system to which the present invention is applied has a configuration in which a plurality of nodes are connected to a communication line. In this system, each node has a central processing unit and the resources controlled by it. This resource is a general term for not only a storage device storing a program file, a data file, etc. but also an image memory of a display unit and the like.

Now, when the central processing unit of any one of the above-mentioned nodes tries to execute a predetermined program, it is necessary to transfer the program or data to the execution memory. The normal data transfer process is performed by repeatedly transferring the minimum unit data group. In the present invention, the minimum unit is called a page, and a program file or a data file required for executing an actual process is called an object. Then, all objects included in all resources of the system are divided into page units, and each page is uniquely specified by a predetermined physical address.

At the generation stage of a predetermined process, so-called mapping is performed in which objects constituting a process to be executed are allocated in the order of execution in segment units, for example. One segment is a unit in which a fixed number of pages are collected. At this time, addresses are assigned to all the segments that make up the process. This address is called a process segment address.

On the other hand, each object is also an aggregate of a predetermined number of segments, in other words, an aggregate of pages having a certain number of times the number of segments.

In the process of incorporating an object into a process, the object is called by filename. In the system of the present invention,
Each node has a table in which the physical addresses of the objects included in the resources of its own node are assigned to the pages that form the objects, and the file names and the physical addresses are associated with each other. This physical address is called an object page address.

Therefore, in the process of generating the process, the physical address of the first page of the segment is listed for each segment of the object included in the resource of the own node. Similarly, for the objects of other nodes, inquire about them before executing the process and make a similar list.

Thus, in the generation stage of the process, a conversion table is created in which all the process segment addresses of the process and the object page addresses of the objects are associated with each other and can be translated. During execution of the process, each conversion table is referred to and each page of the objects of all nodes is directly accessed using the object page address obtained as a result.

As a result, all resources included in other nodes of the system can be accessed in page units at the same high speed as the resources included in the own node to execute the process.

Also, during the execution of the process, for any part of the process, given data to other nodes, that is, the second nodes, including some or all of the conversion table created previously necessary for executing the process. Can be transferred so that the execution subject of the subsequent process can be switched.

In other words, the address translation is performed uniquely in the entire network.

[Example] (Example of node configuration) Fig. 3 is a block diagram showing an example of a node suitable for implementing the inter-node access method in the computer network system of the present invention.

The node 1 includes a network adapter 3 that controls the exchange of signals between the communication line 2 and an internal circuit, an address translator 5 that is connected to the network adapter 3 via a bus line 4, and a central processing unit ( CPU) 6, a main memory 7, a disk 8, a display device 9, and other devices 10 and the like.

The central processing unit 6 comprises a microprocessor and its peripheral circuits for controlling the operation of the entire node.

In addition, the main memory 7, the disk 8 and other devices 10
Each of these is a device for reading or writing data in a memory provided therein by the central processing unit 6 and performing respective operations. The display device 9 includes a graphics memory 9a, a graphics CPU 9b, and a cathode ray tube display 9c.
The data written in the graphics memory 9a is read and output as a video signal to the cathode ray tube display 9c, or an input signal from a keyboard (not shown) is written in the graphics memory 9a.
The central processing unit 6 also reads and writes data in the graphics memory 9a. These devices will be collectively referred to as resources in the present invention.

Further, the network adapter 3 is a circuit having a so-called interface function, but in addition, it has a reference table of object addresses and object names (file names) included in all nodes configuring this system. There is. The entire reference table is stored in the disk 8 or the like, and a part of the reference table is cached and stored in the address translator 5 and the network adapter 3. With this, it is possible to immediately determine which node the object used for the process belongs to from the file name, and the central processing unit 6 can request the node for information regarding the physical address of the object. ing.

The address translator 5 is a storage device that stores a table for address translation, which will be described later.

(Object Address Space) The data contained in the memory of each of the resources 7 to 10 that is read and written by the central processing unit 6 is given a unique address.

In this embodiment, an address is assigned so as to specify these data in byte units, and this address is unique, including that of all the other nodes connected to the communication line 2 in FIG. Have been set.

FIG. 4 is a conceptual diagram showing the address space. This address is an absolute address of hardware and will be called a physical address.

First, the upper 18 bits are used for the node address.
That is, even if 2 ¹⁸ nodes are connected to the communication line,
It is possible to identify. Next, add 24 to the resource address.
Use bits.

Then, the upper 64 bits are used for the address of the program file or data file included in each resource. These 64 bits are set using the resource address 24 bits, the time (year / month / day / hour / minute / second) when the program or the like was created, the access code, and the like.

Usually, before the execution of a process, an operating system is linked with a program file and a data file to perform so-called process generation work. At this time, in order to specify a program file or the like, for example, NODE3 / P
File names such as ROG / KEIRI / KINRI are used. What is designated by this file name is called an object in the present invention, and this object is uniquely specified by the address of the upper 64 bits throughout the entire system.

In addition, the method of transferring the entire objects included in other nodes to the memory of the own node at the generation stage of the process is
It requires a large amount of memory and runs significantly slower.

Therefore, in the case of the access method between the nodes in the computer network system of the present invention, during the execution of the process, the data transfer of the required minimum unit is received,
The method of repeating this is adopted. We call this smallest unit a page, which is 2 ¹² 1-byte data (8 bits)
It is assumed that they are individually collected.

In order to specify this page, the upper 84 bits are used, and this address space is configured so that the physical address can be specified up to the level of 1 byte with a total of 96 bits.

The data transfer operation is that by specifying the page address (called the object page address), 2 ¹² bytes of data are automatically transferred continuously from the top data. The size may be set larger or smaller as necessary.

When the process is executed, the object page address is obtained from the process address given to each step of the process when the process is generated, and the object page address is used for access to transfer the necessary data.

A translation table, which will be described later, is created for this address translation, but a unit called a segment is provided on the page to facilitate the handling of the table. One segment consists of 32 pages and consists of 2 ¹⁷ bytes. Therefore, the upper 79 bits can specify the object segment.

As described above, in the access method between nodes in the computer network system of the present invention, during the execution of the process, the object page address is used to directly access and transfer the predetermined data of the objects included in all the nodes. Receive and use.

For this reason, the central processing unit that intends to execute a given process creates a table that can immediately obtain object page addresses for all the objects required to execute the process at the generation stage of the process. .

(Preparation of process) When preparing a process, first, objects such as program files and data files that compose the process are divided into segments and arranged in the order of execution. This work is also called mapping. That is, as shown in FIG. 5, in this work, for all processes, for example, segment 1 of file A (displayed as FLASEGI. The same applies hereinafter), file A
, Segment 2 of file A, segment 1 of file B, ..., And so on, in the virtual space. In this process virtual space, for example, 16
Processes can be assigned.

An address is given to each segment of the process for which mapping has been completed in this way. This address is called a process segment address.

Similarly, the conversion table shown in FIG. 6 is created in order to associate each process segment address in this virtual space with the physical address of the segment forming the actual object. This conversion table is referred to as a process segment address-object segment address conversion table and will be referred to as a P-O table. This table is also provided with a memory so that 16 processes can be created.

This table is a table that can be referenced bidirectionally from both the process side and the object side. Each process segment address is then displayed as the address of the first page of that segment. A specific page included in the segment can be identified by the number of pages from the first page of the segment. Therefore, P-
The O table is a conversion table for each segment,
Address conversion can be performed in page units by conversion.

That is, by using this P-O table and the conversion arithmetic circuit, for example, the object page address corresponding to a specific page in the process segment can be immediately obtained.

The central processing unit (CPU) creates a P-O table based on the file name at the process preparation stage.
Not only the own node but also other nodes are inquired about the physical address and created.

By creating such a conversion table, it becomes possible to access the data contained in the own node and the data contained in other nodes at substantially the same speed during the execution of the process. The process thus prepared is also stored as a kind of object (it is called a process image).

Then, when the process is executed, the P-O table is first loaded into the address translator 5, then the R-P table is searched for space and the process page is allocated, and if necessary, the resources required for executing the process are secured (this Work is called process generation). Next, as shown in FIG. 7, the data transferred to and written in the execution memory page by page is executed. The execution memory corresponds to a register, a graphic memory, a RAM element, or the like used during the execution of the process, and these have unique addresses. This address is called a real address. For example, this execution memory is set to about 2 ²⁴ bytes.

On the other hand, generally, the process virtual empty space is set to 23 ³² bits. Therefore, during the execution of the process, only the data for a part of the process is transferred to the execution memory. Then, the data is rewritten as the processing progresses. Therefore, a conversion table that associates the process address with the address (real address) of the execution memory is prepared. This is shown in FIG.

This is also a table that can be referenced from both the process address side and the real address side, and is created first when the process is created and is updated each time the process memory is rewritten when the process is executed. This table will be called an RP table. Since the real address is determined by the hardware configuration of the node, only one RP table is created for each node. Therefore, the addresses of all execution memories are added to the real address, and the process address is
The addresses of multiple processes will be mixed.

(Execution of Process) In the access method between nodes in the computer network system of the present invention, the central processing unit executes the process as shown in FIG. A flowchart of the execution is shown in FIG. The description will be given with a focus on these figures.

In FIG. 1, the P-O table 13 (described in FIG. 6) is created in the virtual space 11 in the same manner as described in FIG. 5 above, and the process is prepared. Also, at the time of execution (process load etc.), the RP table 12 (described in FIG. 8) is created, and the data for execution of the first part of the process is transferred to the execution memory 14 (described in FIG. 7). It

In this work, the process segment address in the virtual space 11 is converted into an object segment address in the P-O table 13, and data for that segment is stored in the execution memory 14 page by page from the resource of the own node or the resource of another node. It is done by transferring. (Flowchart step of FIG. 2).

When the process is generated in this manner, the execution memory is accessed through the determination step. Normally, the execution memory is accessed in the order of its real address (step), is executed step by step (step), and this is repeated while making a process end judgment (judgment step).

When there is a so-called jump instruction or data reference, and this reference address does not exist in the same page (step; N), the process proceeds to step.

Since the reference address is usually displayed as a process address, the process address is converted into a real address using the RP table (steps, ...). If this real address is on the same page, the program is executed from step to step and step to step.

On the other hand, if the process address is not in the RP table,
That is, if a page photo occurs, P-
The O table is referred to (step), and this is converted into an object page address (step). Using this object page address, access the data contained in the resource of the own node or another node and transfer the data for one page to the execution memory (step),
The contents of the RP table are updated (step).

Here, in the access method between nodes in the computer network system of the present invention, the central processing unit directly accesses the data of another node by using the object page address, and for any part of the process, the execution subject of that part is set. The operation of switching to the central processing unit of another node can be performed.

For example, according to an instruction designating a node which is an execution subject, the entire conversion table necessary for the process of the part is transferred to the node, and the process is generated and executed by the node.

After the process is created at the node, the process continues as if the node started the process from the beginning. During execution of the process, the execution subject can be moved to another node, returned to the original node, or the process can be terminated as it is.

In fact, if the data necessary for executing the process, including the conversion table, is completely transferred, the P-O table can be used as it is even if the execution subject changes, so the process generation work is completed in an extremely short time. And very efficient.

This is a method suitable, for example, in the case where a node having a hardware capable of efficiently performing a specific operation process for a particular operation process is substituted for that part of the process.

In addition, in the arithmetic processing for handling a large amount of data, it may be possible to reduce the data transmission time due to the interposition of the transmission path by letting the node that owns the data perform the arithmetic operation. Even in such a case, changing the execution subject of the process is effective.

"Modification" The access method between nodes in the computer network system of the present invention is not limited to the above embodiments.

The amount of data that makes up one page is as described above.
You may increase or decrease it arbitrarily.

Further, the conversion table may be freely created by defining a data group in an appropriate unit instead of the segment unit. For other conversion tables, if necessary,
The address may be converted more easily, and more tables may be provided.

Each such conversion table is a disk or R
It may be stored in an AM disk or the like, but caching may be performed using a high speed memory in order to speed up conversion.

"Effects of the Invention" In the access method between nodes in the computer network system of the present invention described above, all the resources of all the nodes constituting the system are managed by the central processing unit of a specific node as the resources of its own node. Equally unique and can be handled in a unified manner, and data can be directly accessed in page units.

Therefore, it is possible to configure a so-called network-wide system in which the execution of the process is speeded up, and the resources of each place are operated in real time from the viewpoint of the entire system instead of the node unit.

[Brief description of drawings]

FIG. 1 is an explanatory diagram showing an outline of an access method between nodes in a computer network system of the present invention, FIG. 2 is a flowchart of its operation, FIG. 3 is a block diagram of a node suitable for carrying out the present invention, and FIG. Is a conceptual diagram of an object address space, FIG. 5 is a conceptual diagram of a process virtual space, FIG. 6 is an explanatory diagram of a P-O table, FIG. 7 is an explanatory diagram of an execution memory, and FIG. 8 is an R-P table. FIG. 6 ... Central processing unit, 7-10 ... Resources.

Claims (2)

[Claims] 1. A method for accessing between nodes in a computer system in which a plurality of nodes each having a central processing unit and resources controlled by the central processing unit are connected by a communication line, and all access methods included in all resources of the computer system. After the page is uniquely specified by a predetermined physical address, the physical address of the page to be transferred to the process execution memory and the process address of the process to be executed are mutually corresponded among the pages specified by the physical address. When a central processing unit of the first node of the nodes executes a predetermined process, the process address is converted into a physical address based on the conversion table, and the converted physical table is created. Objects of all nodes using addresses Access method between nodes in a computer system characterized by performing the process while directly accessing the page. 2. During execution of a process, for any part of the process, predetermined data is transferred to a second node, including a part or all of the conversion table necessary for executing the process. The method for accessing between nodes in the computer system according to claim 1, characterized in that the execution subject of the subsequent process is switched.