JPH05167607A - Coupling network - Google Patents

Abstract

PURPOSE:To devise a network suitable for a high speed operation, to be large- sized easily, in which no traffic concentration takes place in internal paths and a copy function is realized by arranging N/2X(N-1) sets of 2-input/2-output unit switches and interconnecting them in a crossing way. CONSTITUTION:For example, seven-stages each comprising two-sets of two-input/ two-output unit switches arranged longitudinally are arranged in the network. Then the connection from output ports of the stage 1 to input ports of the stage 2, from output ports of the stage 3 to input ports of the stage 4, from output ports of the stage 5 to input ports of the stage 6, is implemented similarly. Moreover, the connection from output ports of the stage 2 to input ports of the stage 3, from output ports of the stage 4 to input ports of the stage 5, from output ports of the stage 6 to input ports of the stage 7, is implemented similarly. In the case of the relation of connection in three-dimensional ways, the length of the connection line used to interconnect the unit switches is selected equal entirely.

Description

Detailed Description of the Invention

 [Object of the Invention]

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a connection network for transferring information input from a plurality of input ports to a desired output port, and particularly for performing information communication using fixed-length short packets called cells. The present invention relates to a connection network suitable for a hardware-implemented cell switch used in an ATM communication system.

[0002]

2. Description of the Related Art In recent years, an STM (Synchronous) that secures an information transmission capacity required for communication at the time of call setup.
A transfer mode called Transfer Mode (used in the existing telephone network) is changed to ATM (Asy
There is an increasing expectation and interest in a transfer mode called an “nchronous Transfer Mode”.

The ATM transmits information using short packets of fixed length called cells, and each communication terminal side passes the cell to the communication network as needed, that is, information of the communication network when the communication terminal needs it. It is a transfer mode characterized by using transmission capacity. Compared to STM, ATM is capable of providing a communication terminal with arbitrary information transmission capacity required by the communication terminal, and improves communication efficiency by using the information transmission capacity of the communication network only when the communication terminal requires it. The advantage is that Therefore, the ATM is a communication network that can handle voices, data, images, and the like in a unified manner, that is, B-ISDN (Broadband Integrar).
arated Services Digital Net
It is in the limelight as a basic technology that constitutes work).

In order to construct an ATM network, it is necessary to realize an operation of switching a cell delivered from a plurality of input communication channels to a desired output communication channel according to the route information of the cells, that is, cell switching. . Since the communication terminal uses the information transmission capacity of the communication network when necessary, a plurality of cells are simultaneously directed to the same output communication path, that is, a conflict occurs when cell switching is performed. If a conflict occurs, transfer one of the conflicting cells to the output channel,
The remaining cells are temporarily stored until there is a vacancy in the output communication path, and output when the vacancy occurs. In order to realize this function, the cell switch is composed of a buffer that temporarily stores cells and a coupling network that transfers the cells to a desired output communication path.

As is well known, in the ATM network, communication is performed based on a kind of connection called a virtual channel (VC). When communication is started between communication terminals, a VC is set between those communication terminals. At this time, for each transmission line in the ATM network through which the above-mentioned VC passes, the VCs set up to that time on the transmission line
It is necessary to confirm that the sum of the capacities occupied by and the capacities of the VCs to be newly set do not exceed the transmission capacity of the transmission line. As a result, it is possible to prevent the traffic having a capacity larger than the transmission capacity from flowing into each transmission line and deteriorating the cell transfer quality (cell discard rate, cell delay).
If the sum of the capacities of the VCs set on the transmission path exceeds the transmission capacity of the transmission path by newly setting the VC, the VC to be newly set is not set and a call loss occurs. deal with. The operation for preventing the VC from being set on the transmission line beyond the transmission capacity of the transmission line is called call admission control. Here, the capacity of a VC is generally defined by a statistic defined in a relatively long time, such as the number of cells belonging to the VC that arrive within a fixed time of a predetermined length. It should be noted that in an extremely short period of time, cells having a capacity larger than the transmission capacity may concentrate on the transmission path (that is, a conflict). As described above, the conflict is avoided by temporarily storing the cell in the buffer in the cell switch.

The VC is also set on the cell switch. FIG. 11 shows how the VC is set on the cell switch. As shown in the figure, the combination of VCs at the input port and the combination of VCs at the output port differ depending on the cell switching. Therefore, in the call admission control, it is necessary to grasp at least the total sum of the capacities of the VCs set in the respective input / output ports of the cell switch.

Even in the internal path of the cell switch, it may be necessary to grasp the total capacity of the VCs. 12
Shows an example of a VC path set on a cell switch when a multi-stage connection network known as a Banyan network is used as the cell switch.
In the figure, the paths between the unit switches in which VC is set between the input / output ports are shown by solid lines, and the paths between the unit switches in which VC is not set are shown by broken lines.

The route of the VC inside the unit switch is further indicated by a broken line. As is well known, in the Banyan network, for example, as shown in the same figure, the traffic is concentrated on the internal route, and the sum of the VC capacities set at the input / output ports is the transmission capacity at the input / output ports. Even if it does not exceed VC, there is a possibility that VC will be set beyond the capacity of the internal path. In such a case, it is necessary to manage the transmission capacity not only for the input / output port but also for the internal path of the cell switch by the call admission control. However, when the transmission capacity is managed also in the internal path of the cell switch in this way, the calculation processing capacity required for call admission control becomes large (especially when realizing a large-scale cell switching system), If you try to realize it, the cost will increase. Therefore, it is desirable that the cell switch has a configuration in which traffic concentration does not occur in the internal route. If a cell switch that does not cause traffic congestion on the internal route is used, the call admission control only needs to manage the transmission capacity only at the input and output ports of the cell switch, reducing the amount of calculation required for call admission control. can do.

The traffic concentration on the internal route of the cell switch appears as a conflict on the internal route of the connection network in a short time. If we can guarantee that no conflicts will occur in the internal routes of the connection network, it means that
It corresponds to ensuring that traffic concentration does not occur inside the cell switch. Cell switches having a configuration in which traffic concentration does not occur on the internal route have been studied so far so as to ensure that no conflict occurs on the internal route of the connection network. Such cell switches can be roughly classified into two classes. The first class is as follows:

In order to ensure that no conflict occurs in the internal route of the connection network, the cell transfer speed is increased in the internal route of the connection network. Specifically, in an N-input N-output connection network, the cell transfer rate from a certain input port to a certain output port is set to N times the cell transfer rate at the input / output port, and the cells are input to the connection network. All cells that can be transferred. Since the cell transfer rate inside the connection network is set to N times as high as that at the input / output port, a situation occurs where cells are transferred from a plurality of input ports at the output port at the same time, that is, a conflict occurs. To avoid this conflict,
A buffer is provided at the output port of the connection network. A cell switch of this class is called an output buffer type cell switch. As a cell switch belonging to this class, the knockout switch proposed by Bell Labs in the United States is particularly famous. In this class of cell switch, a bus structure is generally used as a connection network. Therefore, in the cell switch of this class, it is easy to realize the copy function of transferring the cell input from one input port to the plurality of output ports.

However, in the cell switch belonging to this class, the cell transfer rate of the connection network and the cell input rate of the output buffer are N of the cell transfer rate at the input / output ports, where N is the number of input / output ports. Doubled. Therefore, even if a cell switch of a slightly larger scale (for example, 16 inputs and 16 outputs) is considered, the cell input of the connection network and the output buffer used in the cell switch becomes very high speed, which is very difficult to realize. The second class is as follows.

A buffer (this is called an input buffer) is provided at each input port of the coupling network, and a cell is output from cells stored in each input buffer before the output of cells from each input buffer is started. Perform the operation to select. A functional element that performs this selection operation is called a competition control unit. The contention control unit receives the route information of the cells stored in the input buffer from each input buffer, and for each output port, the cells heading for the output port are selected from the cells stored in the input buffer at most. Select one by one. At this time, at most one cell is selected in each input buffer. Each input buffer is notified of which cell has been selected, and the selected cell is output from the input buffer. Here, if the connection network is so-called non-blocking, the cells output from the respective input buffers are always transferred to different output ports, so that no conflict occurs in the internal path of the connection network. As a result, traffic concentration does not occur in the internal route of the combined network, and it becomes unnecessary to manage the VC capacity of the internal route of the combined network during call admission control. A cell switch of this class is called an input buffer type cell switch. As a switch belonging to this class, the 3-phase algorithm proposed by Bellcore in the United States is particularly famous. Crossbar switches, batcher-banyan networks, and the like are known as non-blocking connection networks.

In the input buffer type cell switch, the cell output rate of the input buffer and the cell transfer rate of the connection network may be the same as or about twice as high as the cell input rate of the input buffer. 600 Mbp
This is advantageous when realizing a high-speed cell switch that exceeds s. However, the input buffer type cell switch known so far has the following drawbacks, and it is very difficult to apply it to a high speed cell switch.

First, the non-blocking connection networks known so far are not suitable for high-speed operation. The crossbar switch needs to broadcast a signal input from one input port to all output ports because of its configuration. Generally, when the broadcast operation is performed, the fanout of each logic element becomes large, which makes it unsuitable for high-speed operation. Therefore, the crossbar switch is not suitable for the fast cell switch.

In the Batcher-Banyan network, the fanout of logic elements can be suppressed to a small level. But,
Its structure is very complex as is well known.
Further, from the operating principle, it is necessary to match the timing of starting inputting cells from each input port in all input ports. This corresponds to the need to operate the entire Batcher-Banyan network with the bit phases aligned. In order to operate the entire Batcher-Banyan network with matching bit phases, it is necessary to distribute the bit clock while controlling the phase throughout the Batcher-Banyan network. From these points, it is very difficult to implement a batcher-banyan network that operates at high speed.

Next, the competition control unit is complicated. The contention control unit has a complicated function of selecting, from the cells stored in the input buffer at a certain time, one cell that outputs at most one from each input buffer so that one cell goes to each output port at most. Is running. The amount of calculation required for this function increases exponentially as the number of input buffers handled increases, that is, as the scale of the cell switch increases. Where the copy function,
That is, considering the function of outputting cells input from one input port to a plurality of output ports, the amount of calculation is further increased. Therefore, it is unlikely that the contention control unit is large-scaled, which makes it difficult to scale up the input buffer type cell switch. Also, the contention control unit realized by hardware is very complicated as well known. Therefore, the competition control unit is not suitable for high-speed mounting.

By the way, there is a multi-stage connection network called a cross network. According to this system, a relatively small non-blocking connection network can be connected in three or more stages to obtain a large-scale connection network. Therefore, it is possible to construct a non-blocking connection network by connecting a relatively small-scale crossbar switch or batcher-banyan network by this method. However, in the non-blocking connection network by this method, the wiring between the stages is complicated and the number of wirings is larger than the number of input / output ports. Therefore, it is difficult to implement a high-speed cross network. Moreover, since the connections between stages are complicated and the number of connections is large, it is difficult to expand the scale. Further, since there are a plurality of paths connecting the arbitrary input / output ports inside the connection network, it becomes necessary to perform routing for each VC inside the connection network, and eventually the VCs set on the internal path of the cell switch are It becomes necessary to manage the capacity.

As described above, in the cell switch which has been proposed so far without traffic concentration on the internal route, high-speed cell switching such as cell switching at a speed of 600 Mbps or more can be easily realized and the scale can be easily expanded. Therefore, there was no cell switch that had the desirable property that the copy function was easy to implement.

[0019]

As described above,
With a cell switch that does not have traffic concentration on the internal route,
High-speed cell switching can be easily realized, the scale can be easily expanded, and the copy function can be easily realized.
There was no cell switch with the desired properties. The present invention has been made in view of the above points, and an object of the present invention is to provide a connection network capable of forming a cell switch having the above-described desirable properties. [Constitution of Invention]

[0020]

Information delivered from one or a plurality of input ports among N input ports (N is a natural number) is transmitted to a desired one or a plurality of output ports among N output ports. There is a connection network that outputs to the output port of the unit, and the connection network is a unit switch with k inputs and k outputs (k is a natural number).
Is configured by arranging N-1 stages having N / k
The input ports of the unit switches included in each stage are numbered as I (0), I (1) ... I (N-1) from the top, and the output ports of the unit switches included in each stage are ordered from the top. O (0), O (1) ... O (N-1) are numbered, and i * 2-1 (i is a natural number, 1 ≦ i ≦ N / 2−1) th stage from the input port side. The output port O (j) of the stage (j is 0 or a natural number, 0 ≦ j ≦ N−1),
The input port I ((j + 1) mod (N-1)) of the i * 2 stage is connected from the input port side, and the output port O ((j + 1) mo of the i * 2 stage is connected.
d (N-1)) is connected to the input port I (j) of the i * 2 + 1th stage.

[0021]

According to the present invention, the self-routing function, that is, the function of naturally guiding information to a desired output port by analyzing and switching the route information included in the information received by each unit switch independently , And the wiring between stages is equal to the number of input / output ports N of the coupling network and is simple, and it is possible to realize a coupling network having desirable properties for high-speed implementation and large scale.

Further, in the coupling network according to the present invention, the information transfer rate of the unit switch and the information transfer rate of the wiring between the stages are twice the information transfer rate of the input / output ports regardless of the number of input / output ports. As a result, a non-blocking connection network is provided. Therefore, according to the present invention, it is possible to provide a cell switch that is suitable for high-speed operation as compared with a bus or cross network, is easy to scale up, and has no traffic concentration on the internal route. .. Further, the connection network according to the present invention has a fanout number of one of the output ports of the unit switches forming the connection network, and therefore is suitable for high-speed operation as compared with the crossbar switch.

In addition, in the cell switch using the connection network according to the present invention, it is possible to disperse the contention controller in each unit switch, which is a problem in the input buffer type cell switch. , Competitive control unit is faster /
It does not become a bottleneck for large scale. Moreover, the cell switch using the connection network according to the present invention can easily realize the copy function.

[0024]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described in detail below with reference to the drawings. (Example 1)

FIG. 1 shows an embodiment of the present invention in which 8 inputs and 8 inputs are used.
The output connection network is shown. In the figure, 11, 12, ...
-74 is a 2-input 2-output unit switch. As shown in the figure, the connection network of this embodiment is constructed by arranging seven stages in which four unit switches each having two inputs and two outputs are arranged vertically and connecting them as follows. ..

In order to describe the connection relationship, I is input to the input ports of the unit switches belonging to each stage in order from the top.
(0), I (1), I (2), ..., I (7) are numbered, and O (0), O (1), O (2), ...
.., numbered as O (7).

Here, the output port of stage 1 is connected to the input port of stage 2, the output port of stage 3 is connected to the input port of stage 4, and the output port of stage 5 is connected to the input port of stage 6 in the same manner. To be done. The connections between these stages are specifically as follows. That is, O (0) becomes I (1) and O (1) becomes I (1).
(2), O (2) to I (3), ..., O
Connect (6) to I (7) and O (7) to I (0).

The output ports of stage 2 to the input ports of stage 3, the output ports of stage 4 to the input ports of stage 5, and the output ports of stage 6 to the input ports of stage 7 are connected in the same way. It The connections between these stages are specifically as follows. That is, O (1) becomes I (0) and O (2) becomes I
(1), O (3) to I (2), ..., O (7)
Is connected to I (6) and O (0) is connected to I (7).

FIG. 2 is a conceptual diagram showing the connection relationship between stages.
Shown in. When the connection relationship is three-dimensionally shown in this way, it can be understood that the connection lines connecting the unit switches can all have the same length. More generally, the connection network according to the invention is described as follows.

The connection network of N inputs and N outputs according to the present invention is 2
N stages with N / 2 input / output unit switches
-1 is arranged side by side. In order to describe the connection between the unit switches, the input ports and output ports of the unit switches included in each stage are numbered as follows. The input ports of the unit switches included in each stage are I (0), I (1), ..., I (N-1) in order from the top.
And number. Further, the output ports of the unit switches included in each stage are sequentially O (0), O (1), ...
., Numbered as O (N-1) (see FIG. 3)

By using this number, the connection between each stage can be described as follows. That is, i * 2-1 (1
≦ i ≦ N / 2-1) Output port O of the stage
(J) (0 ≦ j ≦ N−1) is connected to the input port I ((j + 1) mod (N−1)) of the i * 2nd stage to output the i * 2nd stage. Port O ((j + 1)
mod (N-1)) and the input port I (j) of the i * 2 + 1th stage. FIG. 4 shows an example in which a 4-input 4-output coupling network is constructed in this manner.

By the way, the above-mentioned coupling rule can be applied without particularly limiting the scale of the unit switch to 2 inputs and 2 outputs. For example, a switch with M inputs and M outputs may be considered as a unit switch. In this case, one stage includes N / M unit switches. The number of stages remains N-1. However, since the scale of the unit switch is not particularly relevant to the effectiveness of the present invention, the description will be given here by limiting the scale of the unit switch to 2 inputs and 2 outputs. A self-routing method in the connection network which is an embodiment of the present invention will be described below.

In FIG. 5, an input port having an even input port number, that is, a unit network belonging to stage 1 in the case of a connection network composed of unit switches of 2 inputs and 2 outputs as shown in FIG. Shows the path through which the information input from the input port located in the above is transferred to each output port inside the connection network. In the figure, the paths between the unit switches through which the information input from the input port passes are shown by solid lines, and the paths between the unit switches that do not pass are shown by the broken lines. The path of the information inside the unit switch is shown by a broken line.

Further, in FIG. 6, the input ports having odd input port numbers, that is, the unit switches belonging to stage 1 in the case of a connection network composed of unit switches of 2 inputs and 2 outputs as shown in FIG. The route in which information input from the input port located below is transferred inside the connection network toward each output port is shown. In the same figure, as in FIG. 5, paths between the unit switches through which the information input from the input port passes are shown by solid lines, and paths between the unit switches that do not pass are shown by broken lines. Also, the path of the information inside the unit switch is shown by a broken line.

The information input from the input port of the unit switch 12 includes route information in the information. The route information is (AB) mod N (even input port) (BA) mod N (odd input port) where A is the port number at which the information is output and B is the port number at which the information is input. It is the difference between the output port number and the input port number defined. The bit length of the route information thus defined is logN (base is 2).

In the unit switch of one embodiment of the present invention, the route information of the input information is observed, and if the route information is not 0, the information is crossed, that is, on the upper side in FIG. Information input to the input port is transferred to the lower output port, information input to the lower input port is transferred to the upper output port, and the route information is decremented. If the route information is 0, the information is straight, that is, the information input to the upper input port in the figure is output to the upper output port, and the information input to the lower input port is output to the lower side. Transfer to and from the port. At that time, the route information is not changed. It is necessary to decrement the route information here, but this can be easily realized by a full adder or a counter.

When the input information is switched by each unit switch according to the above-mentioned algorithm, FIG. 5 and FIG.
, And the information is transferred to the desired output port. This can be explained as follows.

As described above, the route information included in the information input to the connection network according to the embodiment of the present invention is the port number of the input port to which certain information is input and the output port to which the information is output. This is the difference information of the numbers. According to the definition of the route information and the topology of the connection network, the operation of decrementing the difference every time switching to the cross side is repeated until the difference becomes 0, and then the information is always transferred to the straight side, which is desirable. Can be transferred to the output port of.
Regarding the information input from the even-numbered input ports and the input port 2 in FIG. 5, the transfer within the connection network will be described in more detail. First, the transfer of information input from the input port 2 and output to the output port 0 within the connection network will be described.

By definition, the route information of the information input from the input port 2 and output to the output port 0 is 6. The information is switched to the cross side by the unit switch 12, and at the same time, the route information of the information changes to 5.

When switched to the cross side, the information is transferred to the upper input port of the unit switch 23. Similarly, the unit switch 23 switches to the cross side, and at the same time, the route information of the information changes to 4.

Similarly, the information is the unit switches 33, 4
4, 54, 61, 71, and accordingly, the route information at the time of inputting the unit switch changes to 4, 3, 2, 1, 0. In the unit switch 71, since the route information when input is 0, the information is switched to the straight side and eventually output to the desired output port, that is, the output port 0.

Similarly, information input from the input port 2 and output to the output port 6 will be described in detail. From the definition, the route information of the above information is 4.
Similarly, the information is the unit switches 12, 23, 33, 4
4, 54, and the route information at the time of inputting the unit switch changes to 4, 3, 2, 1, 0 accordingly. In the unit switch 54, since the route information when input is 0, the information is switched to the straight side and transferred to the unit switch 64. After that, the route information is always 0
Therefore, the unit switch 74 also outputs to the straight side, and eventually the information is output to the output port 6. It is clear that the above description holds true for the information input from the input port 2 to any output port.

Furthermore, information input from any input port having an even number can be similarly transferred to any output port. This is because when the information input to the upper input port of the unit switch belonging to an arbitrary stage of the connection network, which is included in the topology of the connection network according to the embodiment of the present invention, is switched to the cross side, It is obvious considering the property of being output to the upper input port of each unit switch and the property that when all unit switches output information to the straight side, the input port and output port of the same number are connected. .. Next, the transfer of the information input from the odd-numbered input ports and the input port 3 in the connection network in FIG. 6 will be described in more detail.

Here, especially, the information transferred from the input port 3 to the output port 0 will be described. From the definition, the route information attached to the information is 3. Information input from the input port 3 is the unit switches 12, 22, 31, 4
1 is transferred, and the route information when the unit switch is input changes to 3, 2, 1, 0. After the unit switch 41,
Since the route information of the information is 0, it is switched to the straight side, transferred to the unit switches 51, 61, 71 and output to the output port 0 which is a desired output port. Similarly, information input from any input port with an odd number can be transferred to any output port. This is because when the information input to the lower input port of the unit switch belonging to an arbitrary stage of the connection network, which is included in the topology of the connection network according to the embodiment of the present invention, is switched to the cross side, It is clear considering the property that it is output to the lower input port of the unit switch of the stage, and that the property that all unit switches output information to the straight side connects the input port and output port of the same number. Is. Next, the non-blocking property of one embodiment of the present invention will be described.

When the connection network, which is an embodiment of the present invention, is used as a switch for circuit switching, that is, when the connection network is used so that the output ports to which the information given to the input ports are directed do not overlap. FIG. 7 shows a transfer path of information within the connection network. In the figure, the paths between the unit switches through which the information input from the input port passes are shown by solid lines, and the paths between the unit switches that do not pass are shown by the broken lines. Also, the path of the information inside the unit switch is shown by a broken line. Further, among the paths between the unit switches, paths where information is concentrated, 401, ..., 412
Is indicated by a thick solid line.

In FIG. 7, the following coupling is realized for the input port and the output port. That is, input port 0 is output port 7, input port 1 is output port 1, input port 2 is output port 2, input port 3 is output port 3, and input port 4 is output port 4.
, The input port 5 is connected to the output port 5, the input port 6 is connected to the output port 6, and the input port 7 is connected to the output port 0.

According to the above-described routing method inside the connection network, the following is established for the route, for example, the route 401 on which the routing result information is concentrated.

First, the information that joins with the information input from the cross side (for example, the information input from the upper input port of the unit switch 12) must be the straight side (for example, the lower input port of the unit switch 12). It can be said that it is being switched to the information input from). This means that the route information of the information input from the straight side is 0. On the other hand, the route information of the information input from the cross side is not always 0 when it is output from the unit switch of interest here. this is,
This will always be true if no information that is input to the switch at the same time goes to the same output port. This is because if the route information of the two pieces of information on the merged path is both 0, that is, those pieces of information go to the same output port.

From the above, one route (eg, 401)
The information concentrated on the
It will be separated in 2). This holds for all unit switches. Therefore, since at most two pieces of information are concentrated on the path between the unit switches, the speed of the information transfer from the input port to the output port in the unit switch and the information transfer between the unit switches can be determined by the present invention. If the transfer rate of the information at the input / output port of the connection network of one embodiment is set to twice, the connection network becomes non-blocking.

Here, although the information may be concentrated on the path connected to all the input / output ports like the unit switch 43 of FIG. 7, for example, information transfer by the unit switch is performed, for example. First, when the information output to the straight side is transferred and then the information output to the cross side is transferred, the transfer speed of the information in the unit switch is set to double, and thus one embodiment of the present invention. Is a non-blocking network.

Further, if each unit switch has a buffer capable of accumulating two pieces of information transferred by the connection network for each input port, the connection network according to one embodiment of the present invention can be operated in a pipeline manner. Is also possible. By operating in this way, the throughput of the connection network can be improved.

It should be noted that if an input buffer is provided at each input port of the coupling network which is an embodiment of the present invention, and further a competition control unit is provided, the above-mentioned input buffer type cell switch is obtained.

Next, it will be explained that traffic concentration does not occur when the connection network according to one embodiment of the present invention is applied to a cell switch. When applied to a cell switch, a VC is set on the connection network which is an embodiment of the present invention, and cells are transferred inside the connection network. The fact that traffic concentration does not occur means that if the transmission capacity is managed only at each input / output port of the connection network, no matter how the VC is set on the connection network, each inside of the connection network will not be affected. This corresponds to the fact that the sum of the VC capacities set in the route does not exceed the transmission capacity of the internal route.

First, of the routes inside the connection network, the route through which the VC to which the cell switched to the straight side by each unit switch belongs passes. The VC that is the longest in this path, that is, the VC that is switched to the straight side by the most unit switches is the VC that is set between the input port and the output port having the same number.

FIG. 8 shows a VC which is set to transfer cells from all input ports to a specific output port (output port 2 in the figure) in the connection network which is an embodiment of the present invention. Indicates the route. In the figure, the paths between the unit switches through which the VC input from the input port passes are indicated by solid lines, and the paths between the unit switches that do not pass are indicated by the broken lines. Further, the route of the VC inside the unit switch is further shown by a broken line.

In FIG. 8, attention is paid to the path from the input port 2 to the output port 2. The VCs from the respective input ports join the route in the unit switches on the route as described below. That is, the VCs transferred from the input port 3 to the output port 2 merge at the unit switch 12, the VCs transferred from the input port 0 to the output port 2 merge at the unit switch 22, and the VCs transferred from the input port 5 to the output port 2 are merged. The transferred VCs merge in the unit switch 32, the VCs transferred from the input port 6 to the output port 2 merge in the unit switch 42, and the Vs transferred from the input port 7 to the output port 2
C merges at the unit switch 52, VCs transferred from the input port 4 to the output port 2 merge at the unit switch 62, and VCs transferred from the input port 1 to the output port 2 merge at the unit switch 72.

That is, from input port 2 to output port 2
The total sum of the capacities of the VCs set on the transmission path between the unit switches on the route reaching the output path only increases as the output port is approached. Here, since the sum of the VC capacities set on the output port is controlled to be equal to or less than the transmission capacity of the output port, the transmission of the output port on the path from the input port 2 to the output port 2 is controlled. It can be seen that the VC that exceeds the capacity is never set. It is clear that this holds for any set of input ports and output ports.

Next, of the paths inside the connection network which is an embodiment of the present invention, the path to which the VC to which the cell switched on the cross side by each unit switch belongs is set will be considered. The VC that is the longest in this path, that is, the VC that is switched to the cross side by the most unit switches is set between the set of the input port and the output port that is the largest route information in the definition of the route information described above. It is VC.

FIG. 9 shows a VC path set so that each input port goes to the output port having the largest route information in the connection network according to the embodiment of the present invention. In the same figure, the route between the unit switches through which the VC input from the input port passes is indicated by a solid line. Note that V
There is no path between the unit switches through which C does not pass. Further, the route of the VC inside the unit switch is further shown by a broken line. The cell input from each input port is decremented in the route information while being transferred through the route shown in the figure, and when the route information becomes 0, the unit switch switches to the straight side. ..

From the figure, V set from each input port
In C, it can be seen that there is no object passing through the path between the same unit switches in the path that is switched to the cross side by each unit switch. Therefore, if at most the same transmission capacity as that of the input port can be assigned to these routes, there will be no traffic concentration on these routes.

From the above consideration regarding the path of the VC switched to the straight side and the cross side by the unit switch, the VC is set in the path between the unit switches exceeding the sum of the transmission capacities of the input port and the output port. I understand that there is nothing to do. By the way, it is general to make a cell switch so that the transmission capacity of the cell at each input port and the transmission capacity of the cell at the output port become equal. Therefore,
If the path between the unit switches and the cell transfer rate in the unit switch is set to twice as high as the input / output port, traffic concentration does not occur in the cell switch using the coupling network according to the embodiment of the present invention. I understand.

By the way, in the combined network which is one embodiment of the present invention, when the transfer rate of cells in the combined network and the unit switch is doubled, traffic concentration does not occur, but Conflicts will occur. Here, a buffer capable of accumulating a plurality of cells is provided at the input part of the unit switch, and further, for cell transfer between the unit switches,
Introducing an operation that confirms that the cell is surely stored in the destination buffer when outputting the cell, that is, flow control, enables conflict avoidance to be distributed to each unit switch, and the conflict control circuit A cell switch having the desirable property of not determining the upper limit of the scale can be realized. In addition, since a buffer is provided at the input of each unit switch, the connection network operates in an asynchronous mode, that is, in an operation mode of the connection network in which it is not necessary to control the start time of cell output for the route inside the connection network. You will be able to operate. As a result, it is possible to obtain a desirable cell switch in which it is not necessary to control the phase of the bit clock over the entire connection network, which has defined the upper limit of the scale in the Batcher-Banyan network. From these, it can be seen that the factor that determines the scale of the connection network according to the embodiment of the present invention is simply the amount of hardware, and the connection network is a connection network suitable for realizing a large-scale and high-speed cell switch. ..

Furthermore, the present inventors have filed Japanese Patent Application No. 1-1358.
If a buffer used in the cell switch disclosed in 19 above, that is, a buffer that performs well-known control as a window control is adopted as an input buffer of a unit switch of a connection network which is one embodiment of the present invention, two buffers are obtained. Even if a conflict occurs at the head cell of the input buffer, it is possible to introduce into the connection network a function of selecting a cell that can be output from a cell other than the head cell of either buffer and outputting it. As a result, it is possible to reduce the adverse effect of the conflict occurring in each unit switch on the throughput characteristic of the connection network, and thus increase the throughput of the connection network. This makes it possible to further increase the maximum load that can be added to each input / output port, which must be expected when performing call admission control.

The inventors of the present invention filed Japanese Patent Application No. 1-135881.
As disclosed in 9, a unit network having a so-called cut-through function, which starts output of all arriving cells before they are written in the input buffer, forms a connection network according to an embodiment of the present invention. With this configuration, it is possible to mitigate the disadvantage of the connection network, that is, the number of stages is larger than that of the Banyan network, so that the time required for a cell to pass through the connection network becomes large. (Embodiment 2) Next, a method for realizing a copy function in a connection network which is an embodiment of the present invention will be described in detail.

FIG. 10 shows an example of the structure of route information for realizing the copy function. In the figure, the case where the connection network which is one embodiment of the present invention has eight inputs and eight outputs is illustrated. This route information is well known as a bit map system, that is, a bit is prepared for each output port as the route information of the cell, and the bit corresponding to the output port that outputs the cell is set to 1 and the other bits are set. This is route information according to the method in which the bit is set to 0. Further, one bit called a routing end bit is attached. From these, the route information in the connection network which is one embodiment of the present invention having the copy function becomes N + 1 bits when the size of the connection network is N inputs and N outputs.

In the combined network which is an embodiment of the present invention, the unit switches that form the combined network having the copy function switch cells according to the algorithm described below. Since this algorithm is slightly different between the final stage (stage 7 in the connection network shown in FIG. 1) and the other stages (stages 1 to 6 in the connection network shown in FIG. 1), they will be described separately. Here, it is assumed that the routing end bit is reset at the time of input to the connection network.

First, the operation of unit switches belonging to stages other than the final stage will be described. When the routing end bit of the cell sent to the unit switch is reset, switching is performed so that the cell is always output to the cross side. At this time, the routing end bit remains reset. Whether or not the switching is performed so as to output to the straight side is determined by the bit assigned to each stage included in the route information of the cell. That is, if the bit is set, it is also output to the straight side. Then, the routing end bit is set for the cells output to the straight side. If the bit in the route information assigned to each stage is reset, the cell is not output to the straight side.

If the routing end bit of the cell sent to the unit switch is set, the cell is output to the straight side only without being output to the cross side. At this time, the routing end bit for that cell remains set.

Next, the operation of the unit switch belonging to the final stage will be described. If the routing end bit of the cell sent to the unit switch is reset, and if the MSB side is set among the bits assigned to the final stage included in the route information of the cell, , Output the cell to the straight side. Further, if the LSB side is set among the bits allocated to the final stage included in the route information, the cell is output to the cross side. At this time, whichever side is output, the routing end bit of the cell shall be set. If neither of the bits assigned to the final stage included in the route information of the cell is set, the cell is discarded by the unit switch of the final stage without outputting the cell to either exit. ..

On the other hand, when the routing end bit of the cell sent to the unit switch is set, the cell is output only to the straight side. At this time, the routing end bit for that cell remains set.

When the cells are routed as described above, the cells are copied at the required positions on the same routes as those shown in FIGS. 5 and 6, and the connection network is an embodiment of the present invention. Transferred inside. Even when handling a copy connection, the VC is set on the switch. In this case, the VC capacity is managed for each output port that is copied and output. As described above, even when the copy connection is realized, basically, the same routing as the case where the copy connection is not realized is performed, so even if the copy connection is realized in the connection network,
Traffic concentration does not occur inside the connection network.

Of the route information, the meaning of the bits observed at each stage is as follows. That is, the bit closest to the MSB is a bit that specifies whether or not the cell is output to the output port having the same number as the input port that has been input, and the bit one LSB side from the bit is the input port that has been input. A bit that specifies whether or not to output the cell to the output port with a difference of 1, ..., The bit farthest from the MSB is an output port whose difference from the input port is N-1. Is a bit that specifies whether or not to output the cell.

Next, in the connection network which is an embodiment of the present invention,
An example of a unit switch used for realizing a copy connection is shown. This is what the present inventors have proposed in Japanese Patent Application No. 1-1.
This is a modification of the 2-input 2-output cell switch disclosed in 35819 in which the functions of window control, asynchronous cell input, and cut-through are realized.

FIG. 11 shows a connection network which is an embodiment of the present invention and is used for realizing a copy connection.
The structural example of the unit switch of input 2 output is shown. In the figure, reference numeral 801 denotes a 2-input 2-output unit switch, 802a,
b is an input communication path through which cells are delivered to the unit switch 801, 803a and b are output communication paths through which cells are output from the unit switch 801, and 804a and b are input communication paths 80.
Cell input means 805 for receiving cells from 2a, b
Reference numerals a and b denote cell storage means for receiving the cells from the cell input means 804a, b and holding the cells once. Reference numeral 808 denotes switch means for receiving the cells from the cell storage means 805a, b and guiding the cells to a desired route. , B are cell output means for receiving cells from the switch means 808 and outputting the cells to the outside of the unit switch, and 806 is cell storage means 8
The transfer control means 807a and 807b for selecting the cells to be output from 05a and b and controlling the switch means 808 are included in the transfer control means 806, and the cell storage means 805 respectively.
A storage unit that holds information about output ports to which the cells held in a and b are directed. Next, the operation of the unit switch shown in FIG. 11 will be described with reference to FIG.

The unit switch 801 is provided for each input communication path 80.
By asserting the ready signal to the cells 2a and 2b, the cell accumulators 805a and 805b provided corresponding to the respective input ports of the unit switch have vacant spaces for holding new cells. it's shown. The transfer source that transfers the cell to the unit switch refers to the ready signal and outputs the cell while confirming that the unit switch 801 that is the transfer destination of the cell can reliably receive the cell. The ready signal is created by the transfer control unit 806 that knows the cell holding state in the cell storage unit.

When a cell is input from the transfer source, the cell input means 804a, b receives the cell, writes the cell in the cell storage means 804a, b and analyzes the route information of the cell, It is determined which output communication path should be directed to, and the transfer control means 806 is notified of the information. The transfer control unit 806 writes the information about the input cell in the corresponding storage unit 807a or 807b.

Here, the route information of the cell is analyzed according to the self-routing algorithm in the unit switch used for the above-mentioned connection network which is one embodiment of the present invention and has a copy function. Done. Hereinafter, the output port of the unit switch 801 to which the output communication path 803a is connected is the output port U, which means the upper output port, and the output port of the unit switch to which the output communication path 803b is connected, is the lower output port. It will be called an output port D. Depending on the route information, it is displayed whether the cell should be output to the straight side, the one to be output to the cross side, or the one to be output to both output ports. , It is possible to determine whether the cell actually goes to the output port U, the output port D, or the output ports U and D. The cell input units 804a and 804 notify the transfer control unit 806 of the information.

In the transfer control means 806, the cell storage means 8
The cell storage statuses of 05a and 05b are held in the storage units 807a and 807b. The transfer control unit 806 includes a storage unit 807a,
A cell to be output is determined based on the information stored in b, the cell to be output is output from the cell storage means, and the switch means 808 is controlled to guide the cell in a desired direction. Here, by the ready signal input from the output communication paths 803a and 803b, it is confirmed that the unit switch 801 sends a cell to a destination to which a new cell can be held, and the output cell is determined. To do. As a result, flow control is performed inside the cell switch. Switch means 8
The cells output from 08 are the cell output means 809a, b.
Through the output communication paths 803a and 803b. Next, a method of determining cells to be output, which is performed by the transfer control means 806 of the unit switch 801, will be described with reference to FIG.

FIG. 12 shows the structure of the storage units 807a, 807b included in the transfer control means 806 in the unit switch 801.
The structure shown in the figure is provided corresponding to each of the two cell storage means 805a and 805b included in the unit switch 801.

The cell accumulating means 805a, 805b is divided into a plurality of cell accumulating areas 905 having a capacity capable of accumulating exactly one cell. Storage unit 807a,
The data b can hold the following four types of information corresponding to each cell storage area 905.

The flag 904 indicates that the corresponding cell storage area 9
Set if the cell was held at 05, reset otherwise. The U-oriented flag 901 is set if the cell held in the corresponding cell storage area 905 is output to the output port U, and is reset otherwise. The D-oriented flag 902 is set if the cell held in the corresponding cell storage area 905 is output to the output port D, and is reset otherwise.

The counter 903 is used to hold the arrival order among the cells held in the respective cell storage means 805a, 805b. The order of arrival between cells is maintained as follows.

When a cell is written in each cell storage means 805, the cell storage area 905 in the cell storage means 805 which does not hold a cell is selected based on the flag 904, and the selected cell storage area 905 is selected. The cell is written to 905. At the same time, the flag 904 corresponding to the cell storage area 905 in which the cell is written is set, and the U-direction flag 901 and the D-direction flag 902 are set / set according to the route information of the written cell.
Will be reset. Further, the counter 903 corresponding to the cell storage area 905 in which the cell is written is cleared, and at the same time, the other counter 903 in which the corresponding flag 904 is set is incremented.

On the other hand, when the cells are read from the respective cell accumulating means 805, the flag 904 corresponding to the cell accumulating area 905 holding the cell to be output, and the U-direction flag 901 or the D-direction flag 902 are reset. .. At the same time, the counter 903 holding a value larger than the value of the counter 903 corresponding to the cell storage area 905 holding the output cell and having the corresponding flag 904 set are decremented. However, here, when a cell in which both the U-oriented flag 901 and the D-oriented flag 902 are set is output, only the flag corresponding to the output port that outputs the cell is reset, and the flag 904 is reset. And the counter is not decremented. As a result, a cell copy is realized by reading the cell held in the cell storage area 905 twice. The cell output from each cell storage unit 805 is determined by the following method. The following four states are defined by the routes to which the cells stored in the respective cell storage means 805 are directed. State 1: No cells are stored in the cell storage unit 805 State 2: All cells stored in the cell storage unit 805 are for output port U State 3: All cells stored in the cell storage unit 805 are output ports D direction State 4: Some of the cells held in the cell accumulating unit 805 are directed to the output port U, and some are directed to the output port D.

In which state each cell storage means 805
There is a flag 901 for U and a flag 902 for D.
It should be noted that you can easily find out by referring to.

The transfer control means 806 refers to the state of each cell storage means 805, and further, based on the ready signal given from the output communication path, each cell storage means 8
The cell output from 05 is determined. This algorithm may be, for example, one disclosed by the present inventors in Japanese Patent Application No. 1-135819. According to this algorithm, the cells can be output from the cell accumulating means 805a and 805a while keeping the order of arrival at the cell accumulating means 805 and keeping the output ports as empty as possible. Here, the cell switch disclosed in Japanese Patent Application No. 1-135819 does not include a copy function.
A flag 901 and a D flag 902 are prepared independently of each other, and when a cell corresponding to the flag is output when both of the two flags are set, the flag corresponding to the output direction is reset. However, if the state of each cell accumulating means is created from the above two flags as described above, the patent application 1
It should be noted that the algorithm for selecting an output cell disclosed in -135819 can be used. (Embodiment 3) A connection network according to an embodiment of the present invention is
The pattern of the substrate on which the LSI is mounted in the case of combining and realizing the unit switches realized as above will be described.

FIG. 13 shows an outline of the pattern on the substrate for realizing the connection between the unit switches of the 8-input / 8-output coupling network which is an embodiment of the present invention. In the figure, the stage 1 and the stage 2 of FIG. 1 are illustrated. In the figure, 11, 12, 13, 14, 21, 22, 23,
Reference numeral 24 is a unit switch, and 1001 and 1002 are through holes that connect the front and back of the substrate.

In the coupling network which is one embodiment of the present invention, the coupling between the unit switches is very simple. Therefore, for example, half of the unit switches of each stage are mounted on one side of one substrate, and the other side is mounted. By mounting the other half of the unit switches on the surface of, it is possible to form a pattern with no wiring stride as shown in the figure. For this reason, it is possible to design a pattern having a desirable property for high-speed packaging, which has very little crosstalk. Moreover, since all wiring can be performed without physically intersecting the wiring patterns, it is possible to adopt a structure in which mutual interference is unlikely to occur.

[0089]

As described above, according to the present invention,
Since it is possible to provide a non-blocking coupling network easily by connecting very simple unit switches, it is suitable for high-speed operation, easy to scale up, and there is no traffic concentration on the internal route. It can provide a cell switch. Further, in the connection network according to the present invention, it is easy to realize the copy function.

[Brief description of drawings]

FIG. 1 is a diagram showing a connection network according to an embodiment of the present invention.

FIG. 2 is a conceptual diagram showing a connection relationship between stages of a connection network which is an embodiment of the present invention.

FIG. 3 is a diagram showing an input port and an output port of a stage constituting a connection network which is an embodiment of the present invention.

FIG. 4 is a diagram showing a connection network according to an embodiment of the present invention.

FIG. 5 is a diagram for explaining a routing method inside a connection network which is an embodiment of the present invention.

FIG. 6 is a diagram for explaining a routing method inside a connection network which is an embodiment of the present invention.

FIG. 7 is a diagram for explaining non-blocking property of a connection network which is an embodiment of the present invention.

FIG. 8 is a diagram illustrating that traffic concentration does not occur in a connection network that is an embodiment of the present invention.

FIG. 9 is a diagram for explaining that traffic concentration does not occur in a connection network according to an embodiment of the present invention.

FIG. 10 is a diagram for explaining a format of route information used in a connection network that realizes a copy function in a connection network according to an embodiment of the present invention.

FIG. 11 is a diagram showing a configuration example of a unit switch used in a connection network which is an embodiment of the present invention.

FIG. 12 is a diagram showing information used to determine an output cell in a unit switch used in a connection network which is an embodiment of the present invention.

FIG. 13 is a diagram for explaining an example of mounting a connection network on a substrate according to an embodiment of the present invention.

FIG. 14 is a conceptual diagram of VC switching by a cell switch.

FIG. 15 is a diagram illustrating traffic concentration on the Banyan network.

[Explanation of symbols] 11, 12, 73, 74 Unit switch

Claims (1)

[Claims] 1. N input ports (N is a natural number and an even number)
A connection network for outputting information delivered from one or a plurality of input ports among the N output ports to a desired one or a plurality of output ports, and a k-input-k-output unit switch ( (k is a natural number) is arranged in N / k columns to form a stage, (N-1) stages are arranged in parallel, and I is connected to an input port of a unit switch included in each stage.
(0), I (1) ... I (N-1), and the output ports corresponding to the numbered input ports of the unit switches included in the stage are O (0), O
(1) ... Numbered as O (N−1), the output port of the i * 2-1 (i is a natural number, 1 ≦ i ≦ N / 2−1) th stage from the input port side of the connection network O (j) (j is 0 or a natural number, 0 ≦ j ≦ N−1),
The input port side is connected to the input port I ((j + 1) mod (N-1)) of the i * 2 stage, and the output port O ((j + 1) of the i * 2 stage is connected.
mod (N-1)) is connected to the input port I (j) of the i * 2 + 1th stage.