US 4,638,475 AGrant
Router Unit and Routing Network for Determining an Output Port by Detecting a Part of an Input Packet
Issue Date:1987-01-20
•6 Claims
•7 Drawing Sheets
Abstract
In a router unit (20) for use in supplying an input packet (PI) from an input port (I) to an output port (O) with reference to a destination bit series included in the input packet, each input buffer (22) divides the destination bit series into a plurality of partial bit series specified by series positions. A predetermined one of the series positions is assigned to the router unit and is represented by a position signal (PS) to select the partial bit series from the predetermined series position. An output buffer (46) is determined in consideration of the selected partial bit series and is loaded with the input packet. The input packet is sent from the output buffer to the output port as an output packet (PO) to protect the input packet from conflicting with another input packet at the output port. A plurality of the router units may be included in a plurality of stages each of which is given the position signal different from that of the other stages. Alternatively, each router unit may comrise a plurality of code substitution circuit for substituting an input port code for the specified partial bit series to specify an originating input port.
Metadata
Assignee
- NEC Corporation
Inventor
- Nobuhiko Koike
Application Information
Application Number:US 5062640
Filing Date:1983-06-21
Priority Date:1982-06-21
Art Unit:263
Classifications
IPC:
H04Q 1104
Field of Search:
37034060;94825.52;825.53
Patent Drawings (7 sheets)
Description
Background of the Invention
This invention relates to a router unit and a routing network for use as a transmission network in an electronic computer system.
An electronic computer system often has a plurality of processors and is put into operation as a multiprocessor system. Such a multiprocessor system is helpful to improve performance and throughput of the electronic computer system.
With the multiprocessor system, each processor partially takes charge of work to be processed and is operable in relation to the other processors. Data transmission is necessary between the processors through a transmission network arranged therebetween. The performance of the electronic computer system is therefore dependent on that of the transmission network.
A conventional transmission network is put into operation in a manner similar to a crossbar exchange system and is convenient because each processor can be connected through the transmission network to another processor. With an increase of the processors in number, the number of circuit elements necessary for the transmission network should inevitably be augmented in the order of N.sup.2 where N represents the number of the processors. The conventional transmission network therefore becomes expensive with the increase of the number N of the processors.
According to a technical paper contributed by Jack B. Dennis et al to SIGARCH Newsletter Volume 8, Number 3, pp. 1-8, for the seventh annual symposium held during May 6-8, 1980 under the title of "Building Blocks for Data Flow Prototypes," a routing network system is proposed as another conventional transmission system for transmitting each packet from an input port to an output port. Such a routing network comprises a plurality of router units arranged into a plurality of stages comprising a preceding stage and the succeeding stage following the preceding stage and therefore may be called a multistage network system. With the multistage network system, it is possible to accomplish performance equivalent to that of the crossbar exchange system. In addition, costs rise up only in the order of (N log N) even when the number N of the processors is increased. The multistage network system is therefore economical as compared with the before-described conventional routing network system.
However, conflicts or collisions unavoidably take place when a plurality of the processors concurrently transmit the packets to an identical processor through the multistage network system. The performance is seriously reduced on occurrence of the conflicts.
Furthermore, each router unit should be different in structure from the other router unit at every stage of the multistage network system because different operation must be carried out at every stage. In other words, each router unit is not interchangeable with one another.
It is preferable that each router unit of the succeeding stage can recognize each input port of the stage situated prior to the succeeding stage.
Summary of the Invention
It is an object of this invention to provide a router unit which is applicable to a multistage network system and which is usable in common to each stage of the multistage network system.
It is another object of this invention to provide a router unit of the type described, which is capable of avoiding conflicts when two or more of processors concurrently transmit packets to a single processor.
It is a further object of this invention to provide a router unit of the type described, which is capable of recognizing each input port.
It is a yet further object of this invention to provide a routing network which is capable of effectively carrying out transmission of packets.
A router unit to which this invention is applicable is responsive to a plurality of input packets supplied through a plurality of input ports, respectively, for distributing the input packets to at least one output port as output packets. Each of the input packets includes a destination bit series representative of a destination and divisible into a plurality of partial bit series placed at bit series positions, respectively.
According to this invention, the router unit comprises first means for producing a position signal indicative of a predetermined one of the bit series positions that is preassigned to the router unit, a plurality of second means responsive to the respective input packets and to the position signal for selecting predetermined ones of the partial bit series from the predetermined positions, respectively, and third means coupled to the respective second means and responsive to the respective input ports for delivering the input packets to the output ports as the output packets with reference to the predetermined partial bit series, respectively.
Brief Description of the Drawing
FIG. 1 shows a block diagram of a router unit according to a first embodiment of this invention;
FIG. 2 shows a block diagram of an input buffer for use in the router unit illustrated in FIG. 1;
FIG. 3 shows a time chart for use in describing operation of the input buffer illustrated in FIG. 2;
FIG. 4 shows a block diagram of a selector for use in the router unit illustrated in FIG. 1;
FIG. 5 shows a block diagram of a decoder for use in the router unit;
FIG. 6 shows a block diagram of an arbiter for use in the router unit;
FIG. 7 shows a block diagram of an output buffer for use in the router unit;
FIG. 8 shows a block diagram of a routing network comprising a plurality of the router units illustrated in conjunction with FIGS. 1 through 7;
FIG. 9 shows a block diagram of a router unit according to a second embodiment of this invention;
FIG. 10 shows a block diagram of a decoder section for use in the router unit illustrated in FIG. 9;
FIG. 11 shows a block diagram of a code substitution circuit for use in the router unit illustrated in FIG. 9, together with an input buffer illustrated in FIG. 9;
FIG. 12 shows a block diagram of an output circuit for use in the router unit illustrated in FIG. 9; and
FIG. 13 shows a block diagram of a routing network comprising a plurality of the router units illustrated in FIG. 9.
Description of the Preferred Embodiments
Referring to FIG. 1, a router unit 20 according to a first embodiment of this invention is for use in a multistage network system which will later be described and comprises a plurality of stages to be operable in response to first through N-th input packets PI. The first through the N-th input packets will be denoted by PI.sub.1, PI.sub.2, . . . , PI.sub.J, . . . , and PI.sub.N. The input packets PI are supplied through first through N-th input ports I.sub.1 to I.sub.N (symbolized by thick lines on the leftmost side of FIG. 1), respectively. The router unit 20 distributes each of the input packets PI as each of output packets PO to one of first through M-th output ports 0.sub.1 to 0.sub.M. The output packets PO on the output ports 0.sub.1 to 0.sub.M will be designated by PO.sub.1 to PO.sub.M, respectively. The number M may be equal to or more than unity. Each of the input and the output ports may be a shift register.
Each input packet PI includes a destination bit series representative of a destination to which each input packet is to be delivered. The destination bit series is divisible into a plurality of partial bit series which can be specified by bit series positions, respectively. Each partial bit series may be of a single bit. Each input packet PI comprises a data series which follow the destination bit series.
The router unit 20 comprises a signal generator 21 for producing a position signal PS indicative of a predetermined one of the bit series positions that is preassigned to the router unit 20. The position signal PS may be produced by selecting one of a plurality of built-in terminals attached to the router unit 20. The selected one of the built-in terminals may be, for example, grounded and is determined in consideration of each stage which comprises the illustrated router unit 20.
The first through the N-th input packets PI.sub.1 to PI.sub.N are memorized in first through N-th input buffer 22.sub.1 to 22.sub.N, respectively. By way of example, description will be made as regards the J-th input buffer 22.sub.J and is similarly applied to any other input buffers without modification.
Referring to FIGS. 2 and 3 together with FIG. 1, the J-th input buffer 22.sub.J is supplied with the J-th input packet PI.sub.J and a J-th transmission control pulse C.sub.J through the J-th input port I.sub.J. The J-th transmission control signal C.sub.J is sent from a stage preceding the illustrated router unit 20, together with the J-th input packet PI.sub.J. In FIG. 3, the J-th input packet PI.sub.J is not received by the J-th input buffer 22.sub.J during absence of the J-th transmission control pulse C.sub.J. Therefore, the J-th input buffer 22.sub.J is disabled during a first duration I. The illustrated J-th transmission control pulse C.sub.J last a second duration II and a third duration III. Therefore, the J-th input packet PI.sub.J is received by the J-th input buffer 22.sub.J the second and the third durations II and III.
In FIG. 2, the J-th input buffer 22.sub.J comprises a buffer memory 24 and a buffer control circuit 25 comprising, in turn, first and second D-type flip flops (will simply be abbreviated to F/F) 26 and 27. Each of F/F's has terminals D, CK, S, R, Q, and Q all of which are known in the art. Each terminal D of the first and the second F/F's 26 and 27 is put in a logic "1" level, as shown in FIG. 2. The terminal Q of the first F/F 26 is connected to the terminal R of the second F/F 27 while the terminal Q of the second F/F 27, to the terminal R of the first F/F 26. In an initial state, the terminal R of the second F/F 27 is kept in the logic "0" level and is therefore locked in a reset state. The terminal R of the first F/F 26 is kept in the logic "1" level and is unlocked. The first F/F 26 is therefore operable in response to the J-th transmission control pulse although the first F/F 26 is reset in the initial state.
Under the circumstances when the J-th transmission control pulse C.sub.J is given to the terminal CK of the first F/F 26, the logic "1" level appears at the terminal Q of the first F/F 26. As a result, the second F/F 27 is unlocked from the reset state. Simultaneously, the logic "1" level is delivered through a first AND gate 28 to the buffer memory 24 as a memory enable signal EN. Responsive to the memory enable signal EN, the buffer memory 24 memorizes the J-th input packet PI.sub.J and successively produces the same. Likewise, the logic "1" level is sent from the first F/F 26 through a second AND gate 29 as a busy signal BY to the J-th input port and is produced through a second AND gate 30 as a ready signal RD to be described later. The busy signal BY is supplied through the J-th input port I.sub.J to the preceding stage.
When the J-th input packet PI.sub.J is delivered towards the output ports 0.sub.1 to 0.sub.M in a manner to be described, an end signal ED is given to the terminal CK of the second F/F 27 to put the second F/F 27 into a set state. In this event, the logic "0" level is sent from the terminal Q of the second F/F 27 to the terminal R of the first F/F 26. Supplied with the logic "0" level to the terminal R of the first F/F 26, the first F/F 26 is put into the reset state. As a result, the second F/F 27 is also put in the reset state because the logic "0" level is given from the terminal Q of the first F/F 26 to the terminal R of the second F/F 27. When the second F/F 27 is put into the reset state, the logic "1" level is supplied from the terminal Q of the second F/F 27 to the terminal R of the first F/F 26. Thus, the buffer control circuit 25 is returned back to the initial state. At the same time, each of the memory enable signal EN and the ready signal RD is given the logic "0" level. The bus signal BY is also given the logic "0" level as shown in FIG. 3 and disappears a fourth duration IV. The J-th input buffer 22.sub.J can receive the next following input packet.
Referring to FIG. 4 together with FIG. 1, the first through the N-th input buffers 22.sub.1 to 22.sub.N transfer the first through the N-th input packets PI.sub.1 -PI.sub.N to first through N-th selectors 32.sub.1 to 32.sub.N, respectively. The J-th selector 32.sub.J exemplified in FIG. 4 comprises a register 33 supplied with the J-th input packet PI.sub.J through the J-th buffer memory 24 (FIG. 4). The register 33 may keep only the destination bit series included in the J-th input packet PI.sub.J and divides the destination bit series into first through L-th partial bit series PB.sub.1, PB.sub.2, . . . , PB.sub.L. As pointed out hereinabove, each of the partial bit series may be of a single bit.
The first through the L-th partial bit series are sent from the register 33 to a local multiplexer 34 operable in response to the ready signal RD (FIG. 2) and the position signal PS described in conjunction with FIG. 1. The local multiplexer 34 selects a predetermined one of the partial bit series PB.sub.1 to PB.sub.L with reference to the position signal PS to produce the predetermined partial bit series PPB.sub.J. Similar operation is carried out in any other selectors to derive predetermined partial bit series PB.sub.1 to PPB.sub.N except PPB.sub.J.
Thus, each of the selectors 32 (suffixes omitted) cooperates with each of the input buffers 22 in order to select each of the predetermined partial bit series PPB.
Referring to FIG. 5 afresh and to FIG. 1 again, the predetermined partial bit series PPB are supplied from the selectors 32 to first through N-th decoders 36.sub.1 to 36.sub.N, respectively. In FIG. 5, the J-th decoder 36.sub.J comprises a decoding unit 38 supplied with the predetermined partial bit series PPB.sub.J. The decoding unit 38 selects one of the first through the M-th output ports that is indicated by the destination bit series. For this purpose, first through M-th request signal lines are extended from the decoding unit 38 for the respective output ports to send first through M-th request signals Q.sub.J1 to Q.sub.JM towards the respective output ports. Each of the first through the M-th request signals Q.sub.J1 to Q.sub.JM is selectively produced from the decoding unit 38 and may be referred to as an output port signal specifying each output port.
First through M-th acknowledgement signals A.sub.J1 and A.sub.JM are sent to the J-th decoder 36.sub.J through first to M-th acknowledgement signal lines from output port side in a manner to be described later, respectively. Each acknowledgement signal is given to the decoder 36.sub.J when the J-th input packet is delivered to any one of the first through the M-th output ports. Each of the first through the M-th acknowledgement signals A.sub.J1 to A.sub.JM is supplied to an OR gate 39 to be transferred as the end signal ED to the second F/F 27 (FIG. 2). The respective acknowledgement signals A.sub.J1 to A.sub.JM and the corresponding request signals Q.sub.J1 to Q.sub.JM are used in pairs on carrying out transmission between the decoder 36.sub.J and the first to the M-th output ports. Signal pairs of the acknowledgement signals A.sub.J1 to A.sub.JM and the corresponding request signals Q.sub.J1 to Q.sub.JM are transmitted through signal lines R.sub.J1 to R.sub.JM, respectively.
Referring back to FIG. 1, each of the first through the N-th decoders 36.sub.1 to 36.sub.J is connected to first through M-th arbiters 41.sub.1, 41.sub.2, . . . , 41.sub.M through the signal lines R.sub.11, R.sub.12, . . . , R.sub.1M ; R.sub.21, R.sub.22, . . . , R.sub.2M ; . . . ; R.sub.J1, R.sub.J2, . . . , R.sub.JM ; . . . ; and R.sub.N1, R.sub.N2, . . . , R.sub.NM. As a result, each of the arbiters 41.sub.1, 41.sub.2, . . . , 41.sub.M is operable in response to the request signals equal in number to N.
Temporarily referring to FIG. 6, a K-th arbiter 41.sub.K is representative of the first through the M-th arbiters 41.sub.1 to 41.sub.M and is connected to the first through the N-th decoders 36.sub.1 to 36.sub.N through the signal lines R.sub.1K, R.sub.2K, . . . , R.sub.NK. The K-th arbiter 41.sub.K comprises a priority encoder 42 for selecting the request signals Q.sub.1K to Q.sub.NK to produce an encoder output signal EC.sub.K. The priority encoder 42 gives a priority to the request signals in a preselected order when a plurality of the request signals are concurrently received at the priority encoder 42. At any rate, one of the request signals Q.sub.1K to Q.sub.NK is allowed to pass through the K-th arbiter 41.sub.K as the encoder output signal EC.sub.K. The encoder output signal EC.sub.K may be called an allowed request signal. Such a priority encoder 42 can be implemented in a known manner. The priority encoder 42 may select each request signal in the order of occurrence of the request signals, unless a plurality of the request signals are concurrently received by the priority encoder 42.
The encoder output signal EC.sub.K is delivered to a decoding module 43 and towards the output port corresponding to the K-th arbiter 41.sub.K, as will presently become clear. The decoding module 43 decodes the encoder output signal EC.sub.K into one of the acknowledgement signals A.sub.1K, A.sub.2K, . . . , A.sub.NK in response to a set signal SET supplied from the corresponding output port when the input packet is set in the corresponding output port.
Let the one acknowledgement signal be supplied from the K-th arbiter 41.sub.K to the J-th decoder 36.sub.J illustrated with reference to FIG. 5. In this event, the J-th decoder 36.sub.J sends the end signal ED to the J-th input buffer 22.sub.J illustrated in FIG. 2. Thus, the input buffer 22.sub.J is put into the initial state for receiving the following input packet.
Further referring to FIG. 1, the first through the M-th arbiters 41.sub.1 to 41.sub.M supply first through M-th multiplexers 45.sub.1 to 45.sub.M with first through M-th encoder output signals EC.sub.1 to EC.sub.M in a manner similar to that illustrated in conjunction with FIG. 6, respectively. Each of the first through the M-th multiplexers 45.sub.1 to 45.sub.M is supplied with the first through the N-th input packets PI.sub.1 to PI.sub.N through the input buffers 22.sub.1 to 22.sub.N.
Each multiplexer 45.sub.1 to 45.sub.M selects one of the input packets with reference to each of the first through the M-th encoder output signals EC.sub.1 through EC.sub.M. The first through the M-th multiplexers 45.sub.1 to 45.sub.M are connected to first through M-th output buffers 46.sub.1 to 46.sub.M, respectively. Each selected input packet is sent from the multiplexer to the corresponding output buffer. Thus, the first through the M-th multiplexers 45.sub.1 to 45.sub.M produce first through M-th selected input packets PI.sub.S1 to PI.sub.SM, respectively.
Referring to FIG. 7, a K-th output buffer 46.sub.K is representative of the first through the M-th output buffers 46.sub.1 to 46.sub.M and is operable in response to a K-th one PI.sub.SK of the selected input packets. The output buffer 46.sub.K comprises an output buffer memory 50 for storing the K-th selected input packet PI.sub.SK and an output control circuit 51 for supplying the K-th input packet to a K-th output port 0.sub.K as a K-th output packet PO.sub.K.
The K-th selected input packet PI.sub.SK is also supplied to the output control circuit 51. A busy signal BSY is supplied from the succeeding router unit to the output control circuit 51 to indicate a busy state of the succeeding router unit. The busy signal BSY takes the logic "1" level during the busy state and otherwise the logic "0" level. Let the busy signal BSY take the logic "0" level.
The output control circuit 51 comprises first, second, and third D type flip flops 56, 57, and 58 which will be abbreviated to first, second, and third F/F's, respectively, and each of which has terminals S, R, D, CK, Q, and Q. The terminals S and D of each F/F are kept at the logic "1" level. An AND gate 59 is connected between the terminal Q of the first F/F 56 and the terminal CK of the second F/F 57. The busy signal BSY is given to the terminal CK of the third F/F 58 and to the AND gate 59 through a NAND gate 60.
In an initial state, the terminals Q of each of the first through the third F/F's 56, 57, and 58 are kept at the logic "0" level. The terminals R of the first and the second F/F's 56 are given the logic "1" level from the terminal Q of the third F/F 58 while the terminal R of the third F/F 58 is given the logic "0" level from the terminal Q of the second F/F 57. This means that the first and the second F/F's 56 and 57 are put into an unlocked state or an enable state while the third F/F 58 is kept locked into a reset state in which no input signal can be received.
Under the circumstances, the K-th selected input packet PI.sub.SK is successively stored in the output buffer memory 50 and is given to the terminal CK of the first F/F 56. The first F/F 56 supplies the logic "1" level from the terminal Q of the first F/F 56 to the output buffer memory 50 as a packet reception signal REC representative of reception of the K-th selected input packet PI.sub.SK.
Simultaneously, the packet reception signal REC is delivered as the set signal SET.sub.K to the decoding module 43 illustrated in FIG. 6. Thus, the set signal SET.sub.K is produced when the K-th selected input packet PI.sub.SK is received at the K-th output buffer 46.sub.K.
Production of the packet reception signal REC makes the second F/F 57 produce the logic "1" level as a transmission control signal C.sub.K through the terminal Q thereof. This is because the AND gate 59 is enabled by the NAND gate when the busy signal BSY takes the logic "0" level. As a result, the transmission control signal C.sub.K is sent through the K-th output port to the succeeding router circuit, like the transmission control circuit C.sub.J illustrated in conjunction with FIGS. 2 and 3.
The logic "1" level is given from the terminal Q of the second F/F 57 to the terminal R of the third F/F 58. Therefore, the third F/F 58 is changed from the locked state into the unlocked state in which the busy signal BSY can be received from the succeeding router unit.
In this event, the K-th selected input packet stored in the output buffer memory 50 is successively delivered to the K-th output port as the K-th output packet PO.sub.K.
Supplied with the busy signal BSY from the succeeding router unit, the third F/F 58 is put into the set state. At the same time, the logic "0" level is sent from the terminal Q of the third F/F 58 to the terminals R of the first and the second F/F's 56 and 57. The first and the second F/F's 56 and 57 are therefore rendered into the reset state. Subsequently, the third F/F 58 is put into the reset state by the logic "0" level supplied from the second F/F 57. Thus, the first through the third F/F's 56 to 58 are returned back to the initial state.
With the output buffer memory 50, it is possible to memorize the next following selected input packet even when the busy signal BSY is sent from the succeeding router unit to the output buffer 46 illustrated in FIG. 7. Accordingly, conflicts can be avoided by provision of each output buffer when a plurality of the selected input packets are concurrently supplied to each output buffer.
Referring to FIG. 8, a routing network is divided into a first partial stage, namely, a preceding stage 61 and a second partial stage 62, namely, the succeeding stage. The preceding stage comprises first and second router units which are depicted at 20.sub.1 and 20.sub.2 and each of which is similar to that illustrated in conjunction with FIGS. 1 through 7. The first and the second router units 20.sub.1 and 20.sub.2 may collectivley be called a first set of the router units. Likewise, the succeeding stage 62 comprises third and fourth router units 20.sub.3 and 20.sub.4 similar to that illustrated in FIGS. 1 to 7. The third and the fourth router units 20.sub.3 and 20.sub.4 may be called a second set of the router units.
The first router unit 20.sub.1 has first and second preceding input ports I.sub.11 and I.sub.12 and first and second preceding output ports 0.sub.11 and 0.sub.12 while the second router unit 20.sub.2, third and fourth preceding input ports I.sub.13 and I.sub.14 and third and fourth preceding output ports 0.sub.13 and 0.sub.14. On the other hand, the third router unit 20.sub.3 has first and second succeeding input ports I.sub.21 and I.sub.22 and first and second succeeding output ports 0.sub.21 and 0.sub.22. Likewise, the fourth router unit 20.sub.4 has third and fourth succeeding input ports I.sub.23 and I.sub.24 and third and fourth succeeding output ports 0.sub.23 and 0.sub.24.
It is assumed that a first packet depicted at PI.sub.1 is delivered from the first preceding input port I.sub.11 to the third succeeding output port 0.sub.23 and that a second packet depicted at PI.sub.2 is simultaneously delivered from the third preceding input port I.sub.13 to the third succeeding output port 0.sub.23. The first packet PI.sub.1 has a first destination bit series specified by zeroth and first bits 0 and 1 and a first data bit series P.sub.1 following the first destination bit series. The zeroth and the first bits of the first destination bit series take the logic "1" and the logic "0" levels, respectively, as indicated in the rectangle. The second packet PI.sub.2 has a second destination bit series of "1" and "0" arranged at the zeroth and the first bits, respectively. The second destination bit series is accompanied by a second data bit series P.sub.2.
The first and the second router units 20.sub.1 and 20.sub.2 placed in the preceding stage 61 are supplied with the position signal of "0". The position signal is for specifying the bit series position to be detected by the preceding stage 61. Therefore, the position signal of "0" indicates detection of the zeroth bit.
It may be mentioned here that the first and the second preceding output ports 0.sub.11 and 0.sub.12 are specified by the logic "0" and the logic "1" levels at the zeroth bit position of the first packet PI.sub.1 and that the third and the fourth preceding output ports 0.sub.13 and 0.sub.14 are specified by the logic "0" and the logic "1" levels at the zeroth bit position.
Likewise, the third and the fourth router units 20.sub.3 and 20.sub.4 placed in the succeeding stage 62 are supplied with the position signal of "1" indicative of the first bit position of each packet given to the succeeding stage, namely, the second partial stage 62. Each packet is delivered to the first or the third succeeding output port 0.sub.21 or 0.sub.23 when the first bit of each packet takes the logic "0" level. On the other hand, each packet is delivered to the second or the fourth succeeding output port 0.sub.22 or 0.sub.24 when the first bit of each packet takes the logic "1" level.
Supplied with the first packet PI.sub.1 through the first preceding input port I.sub.11, the first router unit 20.sub.1 detects the zeroth bit position of the first packet PI.sub.1 in compliance with the position signal of "0" and delivers the first packet PI.sub.1 to the second preceding output port 0.sub.12 because the first packet PI.sub.1 has the logic "1" level at the zeroth bit position. The first packet PI.sub.1 is sent from the second output port 0.sub.12 to the third succeeding input port I.sub.23 of the fourth router unit 20.sub.4.
The fourth router unit 20.sub.4 monitors the first bit position of the first packet PI.sub.1 and assigns the first packet PI.sub.1 to the third succeeding output port 0.sub.23 because the first packet PI.sub.1 has the logic "0" level at the first bit position.
Similarly, the second packet PI.sub.2 is assigned to the fourth preceding output port 0.sub.14 as a result of supervision of the zeroth bit of the second packet PI.sub.2. Thereafter, the second packet PI.sub.2 is delivered to the third succeeding output port 0.sub.23 because the second packet PI.sub.2 has the logic "0" level at the first bit position thereof.
Thus, the second packet PI.sub.2 may be sent to the fourth router unit 20.sub.4 simultaneously with the first packet PI.sub.1. However, no conflict takes place in the fourth router unit 20.sub.4 because the order of the first and the second packets PI.sub.1 and PI.sub.2 are controlled by each arbiter included in the fourth router unit 20.sub.4 and each of the first and the second packets PI.sub.1 and PI.sub.2 is stored in the output buffer. This applies to any other router units.
As mentioned above, decentralized operation is carried out in each router unit to select each output port.
Referring to FIG. 9, a router unit 20.sub.a according to a second embodiment of this invention comprises first through N-th input ports I.sub.1 to I.sub.N and first through M-th output ports 0.sub.1 to 0.sub.M, like the router unit 20 illustrated with reference to FIG. 1. First through N-th input packets PI.sub.1 to PI.sub.N are supplied through the first to the N-th input ports I.sub.1 to I.sub.N to first through N-th input buffers 22.sub.1 to 22.sub.N in the manner similar to those illustrated in conjunction with FIGS. 1 and 2, respectively. The first through the N-th output ports 0.sub.1 to 0.sub.N produce first through M-th output packets PO.sub.1 to PO.sub.M, respectively.
The first through the N-th input packets PI.sub.1 to PI.sub.N are sent through the first to the N-th input buffers 22.sub.1 to 22.sub.N to first through N-th decoder sections 65.sub.1 to 65.sub.N, respectively. Each of the first through the N-th decoder sections 65.sub.1 to 65.sub.N is substantially equivalent to a combination of the selector 32 illustrated in FIG. 4 and the decoder 36 illustrated in FIG. 5.
Referring to FIG. 10, a J-th decoder section 65.sub.J is representative of the first through the N-th decoders 65.sub.1 to 65.sub.N. The decoder section 65.sub.J comprises a register 33 for storing a J-th one PI.sub.J of the input packets to produce first through L-th partial bit series PB.sub.1 to PB.sub.L. Each of the first through the L-th partial bit series PB.sub.1 to PB.sub.L may be of a single bit and is specified by bit series positions, respectively. As is the case with FIG. 4, the first through the L-th partial bit series PB.sub.1 to PB.sub.L are sent to a local multiplexer 34 responsive to a position signal PS indicative of a predetermined one of the bit series positions that is preassigned to the router unit 20.sub.a. The position signal PS is given to the local multiplexer 34 in the manner described in conjunction with FIG. 1. Responsive to the position signal PS, the local multiplexer 34 selects one of the first through the L-th partial bit series to send the one partial bit series to a decoding unit 38. The one parital bit series is decoded by the decoding unit 38 into a request signal Q.sub.J. The request signal Q.sub.J specifies one of the output ports 0.sub.1 to 0.sub.M and may therefore be called an output port signal.
Referring back to FIG. 9, the first through the N-th input packets PI.sub.1 to PI.sub.N are also sent to first through N-th code substitution circuits 66.sub.1 to 66.sub.N through the first to the N-th input buffers 21.sub.1 to 21.sub.N, respectively. By way of example, a J-th code substitution circuit 66.sub.J will presently be described in detail. However, similar operation is carried out in any other code substitution circuits.
Referring to FIG. 11 the J-th code substitution circuit 66.sub.J comprises a code generator 70 for generating a input port code IC.sub.J assigned to the J-th input port. In FIG. 9, the illustrated router unit 20.sub.a has the first through the N-th input ports I.sub.1 to I.sub.N. In order to specify the respective input ports I.sub.1 to I.sub.N, N sorts of input port codes are produced by the first through the N-th code substitution circuits 66.sub.1 to 66.sub.N. For example, let N be equal to four. In this event, four of the input ports can be specified by four sorts of input port codes, such as "00," "01," "10," and "11."
In FIG. 11, the J-th code substitution circuit 66.sub.J is supplied with the first through the L-th partial bit series PB.sub.1 to PBL from the J-th input buffer 65.sub.J. A K-th partial bit series PB.sub.K is included in the partial bit series. In addition, the J-th code substitution circuit 66.sub.J is supplied with the position signal PS indicative of the predetermined bit series position, as is the case with the J-th decoder section 65.sub.J. It is assumed that a K-th bit series position is indicated as the predetermined bit series position by the position signal PS.
The first through the L-th partial bit series PB.sub.1 to PB.sub.L are delivered to first through L-th multiplexers 71.sub. to 71.sub.L to which the input port code IC.sub.J is given, repectively. A K-th multiplexer 72.sub.K of the first thorugh the L-th multiplexers 71.sub.1 to 71.sub.L is supplied with the K-th partial bit series PB.sub.K and the input port code IC.sub.J.
The position signal PS is decoded by a decoder 74 to select one of the first through the L-th multiplexers 71.sub.1 to 71.sub.L, namely, the K-th multiplexer 71.sub.K. For this purpose, the decoder 74 supplies the K-th multiplexer 70.sub.K with a first selection signal SEL.sub.1 indicative of selection of the input port code IC.sub.J instead of the K-th partial bit series. On the other hand, each of the remaining multiplexers is supplied with a second selection signal SEL.sub.2 indicative of selection of each partial bit series.
As a result, the K-th multiplexer 70.sub.K produces as an output code X.sub.K the input port code IC.sub.J substituted for the K-th partial bit series while the remaining multiplexers produces as the remaining output codes the remaining partial bit series, respectively. The remaining output codes are represented by X.sub.1, X.sub.2, . . . , X.sub.L.
The respective output codes X.sub.1, X.sub.2, . . . , X.sub.K, . . . , X.sub.L are memorized in an output register 76 to be proudced as a J-th substituted input packet SP.sub.J.
In FIG. 9, the J-th code substitution circuit 66.sub.J is connected to first through M-th output circuits 77.sub.1 to 77.sub.M and delivers the J-th substituted input packet SP.sub.J to the respective output circuits 77.sub.1 to 77.sub.M. The J-th decoder section 65.sub.J is also connected to the first through the M-th output circuits 77.sub.1 to 77.sub.M. However, the J-th request signal, namely, the output port signal is delivered to one of the first through the M-th output circuits 77.sub.1 to 77.sub.M alone, as described with reference to FIG. 10.
Likewise, each of the remaining code substitution circuits delivers each of the remaining substituted input packets, such as SP.sub.1, SP.sub.2, and SP.sub.N, to the first through the M-th output circuits 77.sub.1 to 77.sub.M. Each of the remaining request signals, such as Q.sub.1, Q.sub.2, . . . , Q.sub.N, is given to a selected one of the output circuits 77.sub.1 to 77.sub.M.
Referring to FIG. 12 together with FIG. 9, a K-th output circuit 77.sub.K is typical of the first through the M-th output circuits 77.sub.1 to 77.sub.M. The first through the N-th request signals Q.sub. to Q.sub.N, Q.sub.J inclusive, are supplied with an arbiter 81 included in the K-th output circuit 77.sub.K. The arbiter 81 is equivalent in structure to that illustrated in conjunction with FIG. 6 and is therefore not described any longer. Any any rate, one of the request signals Q.sub.1 to Q.sub.N is selected by the arbiter 81 to be supplied to a multiplexer 82. Stated otherwise, the one request signal alone is allowed to pass through the arbiter 81 and may therefore be referred to as an allowed output port signal. Herein, let the J-th request signal Q.sub.J be selected by the arbiter 81 and be sent to the multiplexer 82 as the allowed output port signal.
Supplied with the first through the N-th substituted input packets SP.sub.1 to SP.sub.N, the multiplexer 82 selects the J-th substituted input packet SP.sub.J in response to the allowed output port signal. The J-th substituted input packet SP.sub.J selected by the multiplexer 82 is given to a K-th output buffer 83 similar in structure to that illustrated with reference to FIG. 7. Therefore, the output buffer 83 can successively memorize the substituted input packets SP.sub.J to produce each substituted input packet SP.sub.J as the K-th output packet 0.sub.K even when the busy signal BSY is sent from the succeeding router unit to the output buffer 83. Thus, any conflict does not take place in the router unit 20.sub.a.
With the router unit 20.sub.a, the substituted input packets are sent to the respective output ports 0.sub.1 to 0.sub.N. This means that each input port of the substituted input packets can be detected by the succeeding router unit by monitoring each input port code.
Referring to FIG. 13, a routing network comprises four of the router units (depicted at 20.sub.a1, 20.sub.a2, 20.sub.a3, and 20.sub.a4) which are illustrated with reference to FIGS. 9 through 12 and which will be called first, second, third, and fourth router units, respectively. The first and the second router units 20.sub.a1 and 20.sub.a2 are arranged in a preceding stage, namely, a first partial stage 61" while the third and the fourth router units 20.sub.a3 and 20.sub.a4 are arranged in the succeeding stage, namely, a second partial stage 62'.
In FIG. 13, input and output ports and position signals PS equivalent to those illustrated in conjunction with FIG. 8 are represented by like reference symbols, respectively. Each of the first through the fourth router units 20.sub.a1 to 20.sub.a4 is therefore responsive to two input packets for producing two output packets, as are the cases with the first through the fourth router units 20.sub.1 to 20.sub.4 illustrated in FIG. 8. In addition, each of the first and the second router units 20.sub.a1 and 20.sub.a2 in the preceding stage 61' monitors the zeroth bit position of each input packet in response to the position signal PS of "0" while each of the third and the fourth router units 20.sub.a3 and 20.sub.a4 monitors the first bit position of each input packet in response to the position signal PS of "1," like in FIG. 8.
The first and the second router units 20.sub.a1 and 20.sub.a2 carry out distribution to the first and the third preceding output ports 0.sub.11 and 0.sub.13, respectively, when the zeroth bit takes the logic "0" level. Otherwise, the first and the second router units 20.sub.a1 and 20.sub.a2 make delivery to the second and the fourth preceding output ports 0.sub.12 and 0.sub.14, respectively.
Likewise, each input packet is delivered from the first and the third succeeding output ports 0.sub.21 and 0.sub.23 when the first bit takes the logic "0" level. Otherwise, each input packet is delivered from the second and the fourth succeeding output ports 0.sub.22 and 0.sub.24.
Moreover, first, second, third, and fourth input port codes are assigned to the first through the fourth preceding input ports I.sub.11, I.sub.12, I.sub.21, and I.sub.22 respectively, and are assumed to be represented by "00," "01," "10," and "11," respectively.
Let an input packet PI (as shown in FIG. 13) be given to the second router unit 20.sub.a2 through the fourth preceding input port I.sub.14. The illustrated input packet PI has a destination bit series of "01" divisible into the zeroth and the first bit positions, respectively, and a data bit series PD following the destination bit series. As readily understood from the destination bit series of "01," the input packet PI is to be distributed to the second succeeding output port 0.sub.22.
More particularly, the second router unit 20.sub.a2 sends the input packet from the fourth input port I.sub.14 to the third preceding output port 0.sub.13 as a result of detection of the logic "0" level placed at the zeroth bit position. In this event, the input port code is substituted for the logic "0" level of the zeroth bit in the second router unit 20.sub.a2 in the above-mentioned manner. Inasmuch as the input port code is specified by a signle bit of the logic "1" level, the input packet PI is modified into a preceding output packet PI.sub.m having a destination bit series of "11." The preceding output packet PI.sub.m is supplied through the second succeeding input port I.sub.22 to the third router unit 20.sub.a3.
The third router unit 20.sub.a3 sends the preceding output packet PI.sub.m to the second succeeding output port 0.sub.22 because the preceding output packet PI.sub.m has the logic "1" level at the first bit position. In this event, the input port code is substituted for the first bit of the preceding output packet PI.sub.m. Thus, the preceding output packet PI.sub.m is modified into a succeeding output packet PO illustrated in FIG. 13. Inasmuch as the input port code of "1" is assigned to the second succeeding input port I.sub.24 and is identical with the logic "1" level at the first bit position of the preceding output packet PI.sub.m, the succeeding output packet PO is not apparently changed from the preceding output packet PI.sub.m. However, the logic "1" level at the first bit position specifies the second succeeding input port I.sub.22.
Thus, the zeroth and the first bits of the succeeding output packets take a pattern of "11" and shows that the succeeding output packet PO comes from the fourth input port.
With the routing system, a part of each packet is used to indicate an output port to which each packet is delivered and is replaced by an input port code assigned to each input port. Therefore, selection of each output port is individually carried out in each router unit. In other words, decentralized operation is possible in each router unit. In addition, it is possible to detect an originating input port on a receiving side of a packet.
While this invention has thus far been described in conjunction with a few preferred embodiments thereof, it will readily be possible for those skilled in the art to put this invention into practice in various manners. For example, this invention is applicable to a system other than the multiprocessor system. In FIG. 11, the substitution circuit may comprise a decoder for decoding the substituted input packet to deliver the same to one of the output circuits.
Claims
What is claimed is:
1. A router unit for use in a routing network to respond to a plurality of input packets supplied through a plurality of input ports, respectively, and to distribute said input packets to at least one of a plurality of output ports as output packets, each of said input packets including destination bit series representative of a destination and being divisible into a plurality of partial bit series placed at bit series positions, respectively, said router unit comprising: first means for producing a position signal indicative of a predetermined one of said bit series positions that is preassigned to said router unit as a function of the position of said router unit in said routing network; a plurality of second means coupled to respective input ports and said first means for selecting predetermined ones of the bit positions from the respective input packets in response to said position signal to derive predetermined ones of said partial bit series corresponding to said predetermined bit positions from said destination bit series; and third means coupled to the respective second means and the respective output ports and responsive to the respective input packets for selecting said at least one output port with reference to said predetermiend ones of the partial bit series selected by said second means so as to deliver each of said input packets as said output packets to said at least one output port selected by said third means.
2. A router unit as claimed in claim 1, wherein said third means comprises: a plurality of decoding means, each being coupled to the respective second means for decoding each of the predetermined partial bit series into an output port signal specifying each of said output ports to which the input packet is to be delivered; a plurality of arbiters, each being coupled to the respective decoding means and being capable of responding to at least one output port signal for allowing only one output port signal to pass therethrough as an allowed output port signal; and a plurality of delivery means, each being responsive to the respective input packets and said allowed output port signal for delivering one of said input packets as one of said output packets to the output port that is indicated by said allowed output port signal.
3. A router unit as claimed in claim 2, therein each of said delivering means comprises: selecting means responsive to the respective input packets and said allowed output port signal for selecting said one input packet with reference to said allowed output port signal; and supplying means for supplying said one input packet to the output port as said one output packet.
4. A router unit as claimed in claim 1, wherein said third means comprises: a plurality of code generating means for generating a plurality of codes which are assigned to the respective input ports; a plurality of substituting means coupled to the respective code generating means and responsive to the respective input packets and said position signal for substituting the respective codes for the respective predetermined partial bit series to produce substituted input packets in which the respective codes are substituted for the predetermined partial bit series; and a plurality of output means, each being coupled to the respective secodn means and the respective substituting means for delivering one of said substituted input packets as one of said output packet to the output port with reference to each of the predetermined partial bit series.
5. A routing network comprising a first partial network and a second partial network which are coupled together, said first partial network comprising a first set of router units, each being responsive to a first set of input packets supplied through a first set of input ports, respectively, for distributing said input packets of said first set to a first set of output ports as a first set of output packets, each of said input packets of said first set including a destination bit series representative of a destination and divisible into a plurality of partial bit series placed at bit series positions, respectively, said second partial network comprising a second set of router units, each being responsive to a second set of input packets supplied through a second set of input ports, respectively, for distributing said second set of said input packets to a second set of output ports as a second set of output packets, said output packets of said first set being supplied to said second partial network as said input packets of said second set, the output ports of the first set in each of the first set router units being connected to the input ports of different ones of the second set router units; each of the router units of said first set comprising: means for producing a first position signal indicative of a first one of said bit series positions that is preassigned to said each router unit of said first set; means responsive to the input packets of said first set and said first position signal for selecting first ones of the partial bit series from said first bit series position; and first delivering means responsive to said first set of said input packets and said first partial bit series for delivering the input packets of said first set to the output ports of said first set as the output packets of said first set; each of said router units of said second set comprising: means for producing a second position signal indicative of a second one of said bit series positions that is preassigned to said each router unit of said second set and that is different from said first bit series position; means responsive to the input packets of said second set and said second position signal for selecting second ones of the partial bit series from said second bit series position; and second delivering means responsive to the input packets of said second set and said second partial bit series for delivering the input packets of said second set to the output ports of said second set as the output packets of said second set.
6. A routing network as claimed in claim 5, wherein: said first delivering means comprises: first code generating means for generating a first set of first codes assigned to the input ports of said first set; first code substituting means responsive to said first codes, the input packets of said first set, and said first position signal for substituting said first codes for said first partial bit series to produce first substituted input packets having said first codes substituted for said first partial bit series; and first output means responsive to said first partial bit series for delivering said first substituted input packets to the output ports of said first set as the output packets of said first set; said second delivering means comprising: second code generating means for generating a second set of second codes assigned to the input ports of said second set; second code substituting means responsive to said second codes, the input packets of said second set, and said second position signal for substituting said second codes for said second partial bit series to produce second substituted input packets having said second codes substituted for said second partial bit series; and second output means responsive to said second partial bit series for delivering said second substituted input packets to the output ports of said second set as the output packets of said second set.