JPH07225744A - Information processor and arbitration method - Google Patents

Abstract

PURPOSE:To simplify the processing and the device and improve the efficiency for using a transmission line by first issuing the permission of use to a node with a longer time interval in advance when transmission line using requests from plural nodes are overlapped. CONSTITUTION:An arbiter 105 transmits connection preparation request packets to respective nodes and calculates the time required for transfer preparation at nodes A and B, the time for data preparation at the node B and the time required for the node B to use a bus 1310 and to finish the use. Afterwards, these values are set to a subtraction timer inside the node and the timer is started. The time up to the end of transfer processing is calculated from those values. Next, when the permission of transfer is applied to nodes C and D, the time up to the start of data transfer is calculated. Continuously, the time of the timer is compared with the calculated predictive time up to the next data transfer start and when the time up to the transfer end is shorter than the time up to the data transfer start, permission issue processing is started to the request. The processing at the nodes C and D is transmitted parallelly with the processing at the nodes A and B without any data collision.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information processing apparatus which is composed of a plurality of nodes and mainly performs parallel calculation, and an arbitration method in the information processing apparatus.

[0002]

2. Description of the Related Art Conventionally, in an information processing apparatus for connecting a plurality of nodes and performing parallel calculation, a transmission line use arbitration method is a method of detecting a collision on the transmission line, and a separate arbiter is arranged to make a request according to a use request. There was a method of granting permission based on order or fixed priority.

[0003]

However, in the method of detecting a collision on the transmission line in the above-mentioned conventional example, the means for monitoring the state of use of the transmission line by another node, the interval management until the re-request after the collision is detected, etc. However, there is a drawback that the processing and the apparatus are complicated.

Further, in the method of arranging a separate arbiter and giving the permission based on the request order or the fixed priority according to the use request, to the node which permits the use of the transmission path after the end of the data transfer which is currently being performed. If it takes a long time to transmit the usage permission, there is a period in which the data transfer is not performed after the end of the data transfer this time, which causes a problem that the use efficiency of the transmission line is reduced.

[0005]

According to the present invention, a plurality of nodes connected by a shared transmission line, an arbiter for granting permission to use the shared transmission line based on a shared transmission line use request from the node, and the arbiter. And means for transmitting information between each node,
A means for measuring an information transmission time required for information transmission by the information transmission means, a means for specifying a shared transmission path occupancy time in data transfer between requested nodes, and a request for using the shared transmission path from the node in the arbiter. And a means for controlling the time interval from the reception of the shared transmission path use request to the granting of the usage permission based on the information transmission time and the shared transmission path occupancy time. By realizing it, the above problems are solved.

Specifically, when the transmission path use requests from a plurality of nodes are duplicated, the permission has already been issued to use the transmission path first, or the first request node currently using the transmission path. And calculating the time interval until the end of use of the transmission path between the target node and the target node, if the permission transmission time to the requesting second requesting node and / or target node is longer than the above time interval, By issuing the permission to use these nodes with a long time interval in advance regardless of the transmission completion confirmation from the node that issued the permission to use the shared transmission path, the use permission information to the node that issued the permission in advance. The transmission period and the data transfer process between the first nodes can be executed in duplicate.

Further, when exchanging variable-length data as transmission data, the node requests the arbiter to use the shared transmission path, and at the same time informs information such as the length of the data to be transmitted on the shared transmission path. Based on the calculated transmission path occupancy time in data transfer between each node, the time interval until the end of transmission path use between the first requesting node currently using the transmission path and the target node is calculated and the request is made. If the permission transmission time to either one or both of the second request node and the target node is longer than the above time interval, the permission to use is issued to the node with a longer time interval irrespective of the transmission end confirmation, etc. Thus, it is possible to execute the use permission information transmission period to the node that issued the permission in advance and the transmission process between the first nodes. .

[0008]

【Example】

(Embodiment 1) In this embodiment, a plurality of nodes are connected by an optical fiber and a concentrator composed of an arbiter and a star coupler is connected. A plurality of transmission lines are multiplexed on the optical fiber by wavelength multiplexing. In the present embodiment, an arbitration wavelength λ1 which is an information transmission means for arbitration and a data transmission wavelength λ2 which is a shared transmission line. Two wavelengths are multiplexed. In this embodiment, the data length for data transfer is fixed to 8 bytes.

FIG. 1 is a diagram showing the configuration of an information processing apparatus according to the present invention.

Reference numerals 101, 102, 103 and 104 are calculation nodes, 115 is a concentrator for connecting the nodes,
116, 117, 118, 119, 120, 121, 1
Reference numerals 22 and 123 denote optical fibers that connect each node to the concentrator. Demultiplexers 107, 108, 109, and 110 demultiplex the wavelength-multiplexed optical signals input from the respective nodes into an arbitration wavelength and a data transfer wavelength. Reference numerals 111, 112, 113 and 114 denote multiplexers for multiplexing the arbitration wavelength and the data transfer wavelength. Reference numeral 105 is an arbiter. Reference numeral 106 is an optical star coupler.

FIG. 2 is a diagram showing a calculation node.

Reference numeral 101 indicates the node A in FIG. The configurations are the same for the nodes B, C, and D. 201 is a display device, 202 is a keyboard, and 203 is a disk device. These are normal personal computers,
It is a general type used in workstations and the like. Reference numeral 204 denotes a processor in the calculation node, and 209
Display controller 205 based on the programs and data stored in the RAM of
Various information processing is performed by using the IO controller 206. These configurations are general as a general-purpose computer and are not limited by the present invention. Reference numeral 217 is a data bus in the calculation node, 218 is a control bus, and 219 is an address bus. These buses are arbitrated by a bus arbiter in the processor 204 and are used for data transfer executed by the processor 204 and an optical data interface 208 (hereinafter also abbreviated as I / F). Two
Reference numeral 07 is an optical arbiter interface. Reference numeral 208 is an optical data interface. Reference numerals 211 and 213 are laser light emitting elements. 212 and 214 are photodiodes. Reference numeral 215 is a multiplexer for synthesizing laser signals of different wavelengths emitted from 211 and 213. 120
Is an optical fiber connected to the concentrator.
Reference numeral 216 is a demultiplexer that demultiplexes the optical signal input from the concentrator. 220 is a bus use request signal group for the processor 204. 221 is an optical arbiter I / F
And an information transfer signal group between the optical data I / F.

FIG. 3 is an overall address map of the system. In this embodiment, 4 gigabytes of the entire system address space is allocated to four nodes for use.

First, a procedure for transmitting information between nodes will be described.

Here, the case where the node A reads data at a specific address in the RAM of the node B will be taken as an example. In this embodiment, as the data to be transferred, it is possible to read the data of a fixed length of up to 8 bytes at a time.

The processor 204 of node A is a normal RA
As in the case of reading M, the bus right to use is first acquired (this is arbitrated by the bus arbiter of the processor), and when the right to use is given, the read address is the node B.
The RAM address 40000000h is output to the address bus and other additional information is output to the control bus to issue a read request. No device responds to this address in node A.

FIG. 4 is a block diagram of the optical arbiter interface 207. In the optical arbiter interface, like the normal memory controller, the address decoder 401, which constantly monitors the information output to the bus, is output by the processor. It detects that the address is an address other than its own node, and notifies the program operating on the node arbitration control processor 402 of this by an external access detection signal of 405. At the same time, address 4 is set in the address latch register 403.
The control signal latch 404 latches control information such as the read / write request type transfer byte count of 0000000h. The node arbitration processor 402 uses a one-chip macro controller in this embodiment, but this configuration is not limited by the present invention, and may be configured by hardware logic or the like. The node arbitration processor 402
Device select signal 408, register select signal 4
The information latched by the address latch 403, the control signal latch 404, and the like is read by a normal read operation using 06, and the connection destination is determined by comparing with the address map of FIG. The arbitration request packet shown in FIG. 5 for making a request is created, and this is sent to the parallel / serial converter 409.
Write in. The parallel / serial converter 409 converts the written information into serial data and outputs the serial data to the laser diode 211.
In 1, the input signal is photoelectrically converted and output as an optical signal having a wavelength of λ1 which is a wavelength for arbitration to the arbiter 105 through the fiber 215 via the multiplexer 215. This configuration is common to all nodes.

FIG. 6 shows a block diagram of the arbiter unit 105. Reference numerals 601, 603, 605, and 607 denote photodiodes, which are separated by the demultiplexers 107, 108, 109, and 110 in the concentrator 115, respectively.
An optical signal having a wavelength of 1, that is, the arbitration request signal is received and converted into an electric signal. Now, the request signal arrives from the node A and is input to the serial / parallel converter 611. The serial / parallel converter 611 converts the input serial electric signal into a parallel signal and simultaneously notifies the data reception detection signal 622 to the arbitration control microcontroller (ACMC) 621. In this embodiment, 621 ACM
C is a ROM storing a program and R used for processing
Consists of a microcontroller with built-in AM,
However, this configuration is not limited by the present invention.

In the ACMC 621, when the data reception detection signal 1 of 622 is received, the serial / parallel converter of 611 is selected by the device select signal of 619 and sent from the node A through the data bus 620 from the internal register. Read the request packet.
After that, by analyzing this packet and comparing it with the address map of FIG.
It is detected that the request is a connection request to the node B. After that, the transmission path use status flag provided in software in the ACMC is checked, and when it is in the usable status, the flag is set to the busy status, and the connection preparation request packet shown in FIG. Write to serial converter.

The two connection preparation request packets created and written by any of the above use the optical signal of λ1,
It is output to the node A and the node B via the multiplexers 111 and 112.

Again at node A, the fiber 116
The optical signal input by is separated by the demultiplexer 216 into a signal of wavelength λ1 and converted into an electrical signal by the photodiode 212 and input to the optical arbiter interface 207. In the 207 optical arbiter interface, the input signal is converted into a parallel signal by the serial / parallel converter 410, and at the same time the arrival of the packet is notified to the node arbitration control processor 402 by the data reception signal 411. The node arbitration processor of the node A reads the connection preparation request packet from the serial / parallel converter 410 using the device select signal 408 and the data bus 407 and detects that the connection is permitted.
The data transmission / reception request signal 1 is used to instruct the optical data interface 208 to wait for data reception.

As a result, the node A becomes ready to receive data.

On the other hand, in the node B, (the configuration of each node is exactly the same, the node B
The operation of will be described. ) Fiber 117 (11 in FIG. 2)
The optical signal input by 6) is transmitted by the 216 demultiplexer to λ
The signal of one wavelength is separated. This is converted into an electric signal by the photodiode of 212 and input to the optical arbiter interface of 207. The input signal is 4
The data is converted into a parallel signal by the serial / parallel converter 10 and is simultaneously notified to the node arbitration control processor 402 by the data reception signal 411. When this notification is detected, the node arbitration processor of the node B reads the connection preparation request packet from the serial / parallel converter 410 using the device select signal 408 and the data bus 407.
The node bus request signal group 220 is used to request the bus arbitration in the node to the arbiter in the processor 204. At the 204 processor, 217, 218, 21
The bus use of 9 is immediately stopped, and the node use request signal of 220 is used to notify the use permission of the bus to the node arbitration control processor. When the bus use permission is given, the node arbitration control processor reads 8 bytes from the address 40000000h to the optical data interface 208 by using the data transmission / reception request signal group 211 based on the contents of the packet of FIG. Instruct to output to the diode.

FIG. 8 shows the optical data interface section. The data transmission / reception request signal instructs the address driver 805 to read the address 40000000h and the data transfer sequencer 804 to read the 8-byte address.

The sequencer 804 instructs the address driver to drive the address by the signal 807,
Then, the control driver 806 is instructed to drive a control signal such as a transfer size and a read / write signal to the bus, whereby a normal memory read transaction occurs on the bus in the node. A memory controller (not shown) drives data in response to this request and then drives an acknowledge signal. When the data transfer sequencer detects the acknowledge signal, it issues a conversion start request to the parallel / serial converter 801 using the signal 810, and then returns to the idle state.

The serial-converted signal has a wavelength λ which is a wavelength for data transfer by the laser diode L2 (213).
It is photoelectrically converted using the wavelength of 2 and is output to the concentrator 115 via the multiplexer of 215.

In the concentrator 115, the wavelength of λ2 is separated from the input optical signal by the demultiplexer 108 and output to the star coupler 106. The star coupler 106 106 distributes this optical signal and outputs it to each node from A to D.

The optical signal of λ2 is distributed to the node A of 101 by the optical fiber of 116 via the multiplexer of 111.

At the node A of 101, the optical signal of λ2 is separated by the demultiplexer of 216 in FIG. 2, converted into an electric signal by the photodiode of 214, and input to the optical data interface of 208. The following is a description of the processing in the node A with reference to FIGS. 8 and 4. In the node A, as described above, the node arbitration processor 402 has instructed to supply data in response to a read request from the processor. That is, the data transfer sequencer 804 sets the data buffer 803 to drive the data bus. The serial / parallel converter 802 converts serial data into parallel data and outputs the parallel data to the data buffer 803, and at the same time, outputs the data reception detection signal 811 to the data transfer sequencer 804.

The data transfer sequencer instructs the control driver 806 to drive an acknowledge signal after a certain delay in order to guarantee the time when the data bus 217 is driven, and at the same time, instructs the node arbitration processor to send an end packet. Creation and transmission are notified using a part of the data transmission / reception request signal 221.

The node arbitration processor creates a transfer end packet whose header indicates the end of transfer, and sends it to the arbiter, as in the connection request packet transmission described above.

As in the case of the connection request packet, the arbiter 105 receives and interprets this transfer end packet, sets the above-mentioned transmission path use state flag in the usable state, and enters the idle state for accepting the next connection request.

As a result, the data at the address 40000000h of the request destination node B is supplied to the processor 204 of the node A.

Regarding the write operation, the processing is performed in substantially the same operation only by reversing the direction of data transfer.

The same processing is performed in transfer between other nodes.

The above description is based on the assumption that requests for using the shared transmission path do not occur in duplicate in the information processing apparatus of the present invention.

The above operation is shown as a timing chart in FIG. The horizontal axis represents the passage of time, the length of the hexagon in the horizontal direction in the figure represents the time required for the processing, and the arrow represents the flow of the processing. The white hexagons indicate arbitration information, and the shaded hexagons indicate data information to be transferred. The upper part of the node A column and the lower part of the node B column show the internal processing in each node, and the lower part of the node A column and the upper part of the node B column show the processing in the interface section of each node. .

The above description and FIG. 9 correspond as follows. First, the processor read request 90 generated in the node A
1 is transmitted from the interface unit (902), is transferred on the transmission path over a certain period of time (903), and reaches the concentrator. The arbiter in the concentrator notifies that the connection request is accepted, creates a packet instructing to receive data and a packet instructing to send data, and sends the packets to the node A and the node B, respectively (904). Each packet is transferred on the transmission path over a certain time (905, 906). Node A
Then, the transmitted packet is received (907), and waits for the data to be transmitted. The node B receives the transmitted packet (908), reads the data in the node B, and transmits it (910). The data passes through the transmission path (911, 912), reaches the node A, and is internally processed (913). After that, a packet is sent from the node A to notify that the series of data transfer is completed (91
4) The data is transmitted over a fixed time (915), and the termination processing is performed in the arbiter (916).

The numbers in the figure used in the transfer between the node A and the node B used above are commonly used in the following description and the figure.

Next, a case where the use requests of the shared transmission line are duplicated will be described. For comparison, first use FIG.
A case where a conventional arbitration method is used for the information processing apparatus of the present invention, that is, a case where the end of the previous data transfer is confirmed and the next shared transmission path connection permission is issued will be described. The white hexagons indicate the arbitration information that reaches the arbiter first, and the black hexagons indicate the arbitration information that is processed subsequently. The upper part of the transmission line column above the concentrator column and the lower part of the transmission line column below the concentrator column are the shared transmission line λ2 channels, the lower part of the transmission line column above the concentrator column, and the concentrator. The upper part of the column of the transmission line below the column of represents the channel of λ1 which is the arbitration transmission line.

This is the case where the node A processor read request of 901 and the node C processor read request of 1001 are generated almost at the same time.

Request packet 100 from node C
2 is sent slightly later than the request packet 902 from the node A, so the request arrival 904 from the node A is the arrival 1 from the node C in the concentrator.
It occurs before 004. Here packets 902 and 10
Since 02 uses λ1 on both transmission lines, it seems that optical signals overlap and cause interference (90
3, 1003), in this example, the transmission line for arbitration does not cause interference because it connects the arbiter and the node on a one-to-one basis.

At 904, the transmission path is in the usable state, and the arbiter sets the above-mentioned use state flag to "in use", and instructs the nodes A and B to execute the transfer. Data transfer between the nodes AB is performed in the same manner as in FIG. 1
Although the request from the node C arrives at the time of 004, the request is put on hold until the time of 916 when the end packet arrives from the node A because the usage status flag is in use. Thereafter, the transmission path use permission is given to the nodes C and D, and the transfer process is restarted. (Preparation and transmission of connection preparation request packet, connection permission packet 1005, transmission of respective packets 1006 and 1007, reception of respective packets 1008 and 1009, reading and transmission of data 1010, data transmission 1011, 1012, data reception 1013). After that, an end packet indicating the end of the series of data transfer is sent to the concentrator (not shown).

In this case, there is a blank period in which the transmission line is not used until the transmission with the wavelength λ2 is once ended in 912 and is used again in 1011.

Next, the arbitration peculiar to the present invention in this embodiment will be described. In the arbiter according to the present invention, when the power is first turned on or the system is reset after the system is connected, the ACMC inside the arbiter is
Creates a time measurement packet indicating that the header part of the request packet in FIG. 5 is time measurement for each node, and sends it to the node A by the above method. Subsequently, a counter provided inside the ACMC Keeps counting up at regular intervals. In the node A that has received the time measurement packet, as in the case of the connection preparation request packet, the node arbitration processor 40
2 interprets the packet and determines that it is a time measurement packet. Then, a time measurement response packet is created and sent to the arbiter. Upon receiving this time measurement response packet, the arbiter reads the value of the counter. Since this counter is counted for a fixed time, the transmission time of information between the concentrator and the node is calculated based on this value. This time measurement is sequentially performed for each node. As a result, four times corresponding to each node A, B, C, D are measured. In the system of this embodiment shown in FIG. 1, the nodes A, B, and C are equidistant from the arbiter, and packets are transmitted at a time of 10 μs. On the other hand, the node D exists at a long distance and it takes 60 μs for transmission.

In the arbiter according to this embodiment, these values are measured when the system is started up and are stored in the table in the ACMC.

In the system of this embodiment, a common fiber is used as a data transmission line which is a shared transmission line by wavelength multiplexing and an arbitration transmission line which is an information transmission means for arbitration, and the arbiter and the star are used. Since the couplers are installed close to each other, the transmission time measured on the arbitration transmission line is λ2.
It can be used as the transmission time between the star coupler and each node in the data transmission line of the wavelength of. In a system having a structure in which a data transmission line and an arbitration transmission line are separately provided without using wavelength division multiplexing or when an arbiter and a star coupler are separately provided, time measurement is performed individually.

FIG. 11 is a processing timing chart at the time of request duplication according to the present invention. The processing will be described below with reference to this figure.

First, a read request is issued from node A to node B (901). At about the same time as 1001, a read request is issued from node C to node D. After performing internal processing in 1002 and 902, respectively, it is sent to the arbiter via the transmission line (90
3, 1003). In the arbiter, the request from the node A arrives first. The arbiter performs the same processing as in the above-described conventional example in 904 and sends a connection preparation request packet to each node, and at the same time, the time required for 10 μs in 905 and 906 is determined by the contents of the above table, and further determined at system design The time required for data transfer preparation at nodes A and B, which is held in the program as a constant (10 μs (90
7, 908)), and like the previous cases 907 and 908, the time from the start of data transmission to the end of data preparation in the node B, which is definite at the time of system design (data preparation time in this embodiment). 1 μs and the sum of 1 μs from the start to the end of data transmission 2 μs (91
0)), and the transfer time from node B to node A via the star coupler in the concentrator (that is, the transfer time from node B to the concentrator (911)).
And the transmission time from the concentrator to node A (91
20 μs) which is the sum of 2) is added to calculate 42 μs, which is the time required for the data according to this request to finish using the transmission path.
After that, this value is set in the subtraction timer inside the node and the timer is started. It also stores that the nodes currently transferring are A and B.

After that, in 1004, the processing of the request from the node C is started. In the processing of ACMC in 1004, the processing at the time of duplication described above is performed, and then the requesting node and its target node are currently transferring. Check that it is not the node you are going to. This node is A or B
If so, the end process arrives and the next process is performed as in the above-mentioned conventional example (if each node can send a new request packet and process a connection preparation request in parallel with the current process, as long as that is possible. is not). In this embodiment, since the request and the target node are C and D, the following processing which is a unique processing according to the present invention is performed. First, the value of the subtraction timer is read, and the time until the end of the transfer process that is currently being given with permission is completed (at the timing of this embodiment, the above-mentioned process is performed in the data transfer process between the nodes AB, which is being processed first). It takes 42 μs, but 32 μs) is obtained because 10 μs has already passed. Next, referring to the table in the ACMC, when the transfer permission is given to the node C and the node D by the same method as the calculation of the setting value to the subtraction timer, the time until the data transfer is started is calculated. 10 in the example
Processing by ACMC 05: 10 μs, transmission line 1007: 60 μs, time required for data transfer preparation 1 of 1009
0 μs, which is 81 μs in total, which is 1 μs at the time 1010 required to prepare data and start sending it to the shared transmission path.

Subsequently, the time until the end of the current transfer read by the timer is compared with the calculated estimated time until the start of the next data transfer, and the time until the end of the transfer is less than the time before the start of the data transfer. (32μs <81
μs corresponds to this in the case of the present embodiment) immediately starts the permission issuing process for the request. In the opposite case, that is, when the time until the end of transfer> the time until the start of data transfer, the time elapses and the time until the end of transfer becomes shorter,
Wait until the time until the transfer ends <the time until the data transfer starts, and then issue the permission for the request. After that, this value is set in the subtraction timer and activated. As a result, even though the previous data transfer was not completed, 100
The arbiter process 5 is started, and the connection preparation request packets 1006 and 1007 are transmitted. Then node C,
The processing at D proceeds in parallel with the processing at nodes A and B as shown in FIG. 11, and ends at 1013 in FIG. 11 without causing data collision on the transfer path.
As is apparent from the figure, the data transfer between the two sets of nodes can be completed in a shorter time than in the case of the conventional processing of FIG.

Further, in the case of the data writing operation to the memory, the same processing can be performed, and the high performance according to the present invention can be expected.

(Embodiment 2) The first embodiment is the second embodiment.
An example in which the data transmission path in the above embodiment is constituted by an electric bus used in a normal computer system will be given. FIG. 12 shows a block diagram of a system having this configuration.

When realized by the bus structure, the time required for data transmission is common to all nodes, and it is necessary to use a fixed time considering the influence of reflection at the end of the transmission line as the transmission line occupation time. In this embodiment, this is set to 20 μs. The transmission time between each node and the arbiter is 60 μs for the node D and 10 μs for the other nodes as in the first embodiment.

This will be described with reference to FIG.
A read request 901 is issued to node B. At about the same time as 1001, a read request is issued from node C to node D. 1002 and 902 respectively
After being processed internally, is sent to the arbiter via the transmission line. In the arbiter, the request from the node A first arrives (904, 1004). The arbiter performs processing at 904 and sends a connection preparation request packet to each node at the same time as in the first embodiment, and at the same time, the time required at 905 and 906 depends on the contents of the table.
0 μs, node A and node B, which are determined during system design and are held in the program as constants
10 μs (907, 90) required for data transfer preparation
8) Further, as in the case of 907 and 908, the data preparation time in the node B that is definite at the time of system design is 1 μs (91
0), and time 20 μs from the beginning to the end of use of the bus 1310 by the node B (1301; 1 in FIG. 13).
301 is shown in the arbiter column for the sake of convenience))), and the time 41 μs until the data by this request ends the use of the transmission path is calculated. In this embodiment, the time from the start of data transmission to the end of data transmission is included in the bus usage time of 20 μs. After that, this value is set in the subtraction timer inside the node and the timer is started.
It also stores that the nodes currently transferring are A and B.

After that, in 1004, the processing of the request from the node C is started. In the processing of ACMC in 1004, the processing at the time of duplication described above is performed, and then the requesting node and its target node are currently transferring. Check that it is not the node you are going to. This node is A or B
In that case, as in the first embodiment, the next process is performed after waiting for the arrival of the termination process. In this embodiment, since the request and the target node are C and D, the following processing which is a unique processing according to the present invention is performed. First, read out the value of the subtraction timer,
The time (31 μs at the timing of this embodiment) until the end of the transfer process which is currently given permission and is being executed is obtained. Next, referring to the table in the ACMC, when the transfer permission is given to the node C and the node D by the same method as the calculation of the setting value to the subtraction timer, the time between data transfer start is calculated. In the embodiment, processing 1 in ACMC 1005
0 μs, 60 μs on 1007 transmission line, 1 of 1009
A total of 81 μs of 0 μs and 1 μs at 1010.

Subsequently, the time until the end of the current transfer read by the timer is compared with the calculated estimated time until the start of the next data transfer, and the time until the end of the transfer is <the time before the start of the data transfer. Case (31μs <81μ
s corresponds to this in the case of the present embodiment) immediately starts the permission issuing process for the request. After that, this value is set in the subtraction timer and activated. As a result, the arbiter process of 1005 is started and the connection preparation request packets of 1006 and 1007 are transmitted even though the previous data transfer is not completed. After that, the processing in the nodes C and D proceeds in parallel with the processing in the nodes A and B as shown in FIG. 13, and the data from the node D is transmitted on the bus without causing data collision on the transfer path. (13
02) The process ends at 1013 in FIG. As is apparent from the figure, the data transfer between the two sets of nodes can be completed in a shorter time than in the case of the conventional processing of FIG.

(Third Embodiment) In the first and second embodiments, data having a fixed length is considered as the data to be transmitted. This Embodiment The following embodiment is intended to transfer variable-length data in the present invention. The configuration of the present embodiment is the same as that of the first embodiment except for some processing in the arbiter,
It is configured as shown in FIG.

In this embodiment, the node which has issued the request to use the shared transmission path writes the length of the data to be transferred in the request packet and outputs it to the arbiter. Figure 14 below
The processing will be described below.

First, in 901, the node A is changed to the node B
A read request is issued. At about the same time as 100
At 1, a read request is made from node C to node D. At this time, the required data length is specified in the request packet issued from the node. In this embodiment, node A specifies 80 bytes. In the present embodiment, the data length is not limited to a fixed length or less as in the first embodiment and can be arbitrarily specified, and is written in the transfer byte number portion of FIG. 15 showing the structure of the request packet. Be done.

The request packets thus created are internally processed at 1002 and 902, respectively, and then sent to the arbiter via the transmission path. In the arbiter, the request from the node A arrives first. The arbiter performs processing at 904 and sends out a connection preparation request packet to each node, and at the same time, the time required for 10 μs at 905 and 906 is 10 μs according to the contents of the table of the first embodiment. Stored in the program as 10 μs (907,
908), the time to prepare the data at Node B 1 μs (9
10) Further, when the use permission is given to the node AB calculated by a program (not shown) which calculates the transmission path occupation time based on the content of the request packet, the node B
Calculates the time until the data transfer by this request finishes the use of the transmission path by adding the occupation time of the transmission path by the packet sent to the transmission path.

The transmission line occupation time is from when the node B starts transmitting data until when the node A finishes receiving the data. In this embodiment, the data transfer rate is 8 per second.
Since it is a megabyte, it takes 10 μs to send 80 bytes of data (in this embodiment, this calculation is performed by dividing the number of transfer bytes in the request packet by 8 and multiplying the result by 1 μs). 2 from B to the node A via the star coupler on the shared transmission line
Since it takes 0 μs, it becomes 30 μs. That is, it can be seen that this data transfer requires 51 μs to complete the use of the transmission path.

At this time, if the number of transfer bytes becomes large, the time required for the data preparation of the node B becomes long, but in reality, each node reads out the data from the internal memory, writes it to the optical fiber, and requests writing. In the case of, the reading of data from the optical fiber and the supply to the processor are executed in an overlapping manner, so the time required to read or read the beginning of the data, in this embodiment, is the fixed time determined at system design. The value 1 μs may be used as the data preparation time.

After that, this value is set in the subtraction timer inside the node and the timer is started. It also stores that the nodes currently transferring are A and B.

After that, in 1004, the processing of the request from the node C is started. In the processing of ACMC in 1004, the processing at the time of duplication described above is performed, and then the requesting node and its target node are currently transferring. Check that it is not the node you are going to. This node is A or B
If this is the case, the next process is performed after the arrival of the end process as in the above embodiment. In this embodiment, since the request and the target node are C and D, the following processing which is a unique processing according to the present invention is performed. First, the value of the subtraction timer is read out, and the time until the end of the transfer process that is currently being performed by giving permission is obtained (41 μs at the timing of this embodiment). Next, above A
In the present embodiment, when the transfer permission is given to the nodes C and D by the method similar to the calculation of the setting value to the subtraction timer by referring to the table in the CMC, the minimum time until the data transfer is started is calculated. Processing 1 in ACMC 1005
0 μs, 60 μs in 1007 transmission line, 10 μs in data transfer preparation 1009, 1 μs in data preparation 1010
Of 81 μs in total.

As described above, the time required for data preparation varies depending on the transfer data length, but here it is necessary to calculate the minimum time until the data transfer is started to the optical fiber, so the minimum 1 μs is set. to add.

Subsequently, the time until the end of the current transfer read from the timer is compared with the calculated estimated time until the start of the next data transfer, and the time until the end of the transfer is <the time before the start of the data transfer. Case (41μs <81μ
s corresponds to this in the case of the present embodiment) immediately starts the permission issuing process for the request. After that, this value is set in the subtraction timer and activated. As a result, the arbiter process of 1005 is started and the connection preparation request packets of 1006 and 1007 are transmitted even though the previous data transfer is not completed. After that, the processing in the nodes C and D proceeds in parallel with the processing in the nodes A and B as shown in FIG. 14, and ends at the point 1013 in FIG. 14 without causing data collision on the transfer path.

Since only fixed length data is handled in the first embodiment, when transferring long data, it was necessary to divide the data and set the fixed length to be long. The division transfer is complicated, and if the fixed length is set to be long, even short data will be transferred in a long data packet and the efficiency will be reduced. However, in the present embodiment, variable length data is handled. Therefore, the shared transmission path can be used efficiently.

Further, the same processing can be performed in the case of the data writing operation to the memory, and the high performance of the present invention can be expected.

Further, in the present embodiment, the number of transfer bytes is notified as the information regarding the use of the shared transmission path, which is notified to the arbiter at the time of requesting the use of the shared transmission path. However, the data length is determined depending on the type of data to be transferred. when,
It is also possible to adopt a configuration in which the type of data to be transferred is notified.

(Fourth Embodiment) As a fourth embodiment, a method of constructing the request packet of FIG. 15 in the third embodiment as shown in FIG. 16 is conceivable. In this embodiment, a fixed number of bytes are used instead of the number of transfer bytes. In the example, one block is 8 bytes, and the transfer request amount is represented by the number of blocks.

As an effect peculiar to this embodiment, it is not necessary to divide by 8 when calculating the time depending on the number of transfer data in the arbiter part, the processing is simplified and the processing time can be shortened. If the transfer rates are different, the number of bytes for one block may be determined correspondingly.

[0073]

As described above, according to the present invention, in an information processing apparatus in which a plurality of nodes use a shared data transmission path, a means for the usage status of the transmission path, a processing means for data collision, etc. are used. In other words, it is possible to reduce wasteful unused time of the transmission line when the usage requests of the data transmission line are multiplexed, and to speed up the processing in the system. Also, the same effect can be obtained in an information processing device in which the transfer data is variable length data.

[Brief description of drawings]

FIG. 1 is a block diagram of an information processing device according to a first embodiment.

FIG. 2 is a block diagram of a node.

FIG. 3 is a diagram showing a system address map.

FIG. 4 is an optical arbiter interface block diagram.

FIG. 5 is a diagram showing the structure of a request packet.

FIG. 6 is a block diagram of an arbiter unit.

FIG. 7 is a diagram showing the structure of a connection preparation request packet.

FIG. 8 is a block diagram of an optical data interface unit.

FIG. 9 is a read processing timing chart.

FIG. 10 is a processing timing chart when requests are duplicated when the conventional arbitration method is used in the information processing apparatus of the first embodiment.

FIG. 11 is a processing timing chart at the time of request duplication according to the present invention.

FIG. 12 is a block diagram of an information processing device according to a second embodiment.

FIG. 13 is a processing timing chart when requests are duplicated according to the second embodiment.

FIG. 14 is a processing timing chart when requests are duplicated according to the third embodiment.

FIG. 15 is a diagram showing the structure of a request packet according to the third embodiment.

FIG. 16 is a diagram showing the structure of a request packet according to the fourth embodiment.

[Explanation of symbols]

101, 102, 103, 104 node 105 arbiter 106 star coupler 111, 112, 113, 114 multiplexer 107, 108, 109, 110 demultiplexer 115 concentrator 204 processor 205 display controller 206 IO controller 207 optical arbiter interface 208 optical data Interface 209 RAM 210 ROM 211, 213 Laser diode 212, 214 Photodiode 215 Multiplexer 216 Demultiplexer 621 Arbitration control microcontroller (ACMC) 1310 Bus transmission line

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Masato Kosugi 3-30-2 Shimomaruko, Ota-ku, Tokyo Canon Inc.

Claims (9)

[Claims] 1. A plurality of nodes connected by a shared transmission path, an arbiter that grants permission to use the shared transmission path based on a request from the node to use the shared transmission path, and information transmission means between the arbiter and each node. A means for measuring an information transmission time required for information transmission by the information transmission means, a means for specifying a shared transmission path occupation time in data transfer between requested nodes, and a shared transmission path from the node in the arbiter. Information processing including means for controlling a time interval from reception of the shared transmission path use request to permission of use based on the information transmission time and the shared transmission path occupancy time when a plurality of usage requests are received in duplicate. apparatus. 2. The information processing apparatus according to claim 1, wherein the shared transmission path occupancy time is specified by performing a fixed length data transfer between the nodes on the shared transmission path. 3. When the shared transmission path use request is transmitted from a node having a request to the arbiter, the shared transmission path occupancy time is specified by simultaneously notifying the arbiter of information regarding the use of the shared transmission path. Claim 1 characterized by
The information processing device described. 4. The information processing apparatus according to claim 1, wherein the shared transmission line is a star-type coupling network. 5. The information processing apparatus according to claim 1, wherein the shared transmission path is composed of an optical fiber. 6. The information processing apparatus according to claim 1, wherein the information transmission means between the arbiter and each node is constituted by an optical fiber. 7. The information processing apparatus according to claim 1, wherein the information transmission means between the arbiter and each node and the shared transmission path are wavelength-multiplexed and configured by a common fiber. 8. A plurality of nodes connected by a shared transmission line, an arbiter that grants permission to use the shared transmission line based on a request from the node to use the shared transmission line, and information transmission means between the arbiter and each node. An arbitration method in an information processing device including: a arbiter, wherein the arbiter requires a previously measured information transmission means when a first use request and a second use request of a shared transmission path are received in duplicate. Based on the information transmission time and the shared transmission path occupancy time that is measured in advance or specified based on the information notified from the requesting node when requesting the use of the shared transmission path, the use of the shared transmission path is permitted first. A first time until the use of the shared transmission path in the first use request ends and a second time before the start of the shared transmission path use when the second use request is permitted An arbitration method characterized by calculating the respective intervals and permitting the second usage request when the second time is longer than the first time. 9. When the first time is longer than the second time, the second use request is permitted after waiting for the second time to be longer than the first time. Item 8. The arbitration method according to item 8.