JPH0360247A - Self-routing channel - Google Patents

Abstract

PURPOSE:To prevent lowering of the service rate of the outgoing circuit of a switch circuit network by adding plural pieces of subswitches in each channel stage. CONSTITUTION:Basic switch elements 21k-0-a (k=1 to 3, a=0 to 3) as circuits (switch elements) to determine a link to be connected based on a connection destination address given to communication information and rank 1 subswitches 21k-1-a and rand 2 subswitches 21k-2-a which are another switch elements are provided. Since plural subswitches are added in each channel stage, even if a fault is caused in a routing route, the communication information can be transmitted to the switch element of a next stage through the subswitch without transferring the communication information to the other switch in the same stage. Thus, the service rate of the outgoing circuit of the switch network can be maintained at a high level.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention is applicable to an ink connection network used for communication between computer processors or a communication path used for high-speed packet switching. The present invention relates to a self-routing channel that performs distributed control based on hardware.

[Prior Art] Figure I0 shows a configuration example of a well-known switch network as a typical self-routing communication path (number of population lines = number of outgoing lines = 8).
It is a circuit that shows (for example, reference document 1) C, L,
Wu and T., Y., Feng, “Ona
class of multistage in
terconnection networks,
IEEE Trans, Computer
s.

vol, c-29, pp, 694-702. August
(1980)].

In the same figure, 130-0 to 130-7 are incoming lines, 13
1-0~131~7 are outgoing lines, 100-0~100~7
101 is a connection information providing circuit that corresponds one-to-one with an incoming line.
-0 to 101-7 are connection information removal circuits that correspond one-to-one with the outgoing line, and 111-0 to l I3-3 are connection information removal circuits that correspond to the outgoing line on a one-to-one basis. This is a circuit (switch element) that determines the appropriate link.

Here, switch elements 1lk-0 to 11k-3 form a communication line stage k. 121-0 to 124-7 are links that connect connection information provision circuits and switch elements.
-7 was set for k-0 to k-12. The link from one communication path to the next communication path is wired so that any incoming line can be connected to any outgoing line.

Routing route selection from the connection information adding circuit to the connection information removing circuit is performed based on the connection destination address added to the communication information by the connection information adding circuit.

The connection information adding circuit adds a connection destination address to the communication information input from the incoming line, and the connection information removal circuit removes the connection destination address.

The address of the connected outgoing line 131-p is expressed in binary (dz
, d+ , do) (However, P = d z・22+d,
・2i +d, ・2°]. The switch element in the channel stage pays attention to d3-k in this bit string, and if it is 0, it transmits the communication information to the upper output link, and if it is l, it transmits the communication information to the lower output link.

After passing through all communication channels, the communication information reaches the specified connection information removal circuit.

The self-routing communication path described above is easy to control because the routing path is uniquely determined, but if a failure occurs at even one point within the routing path, information transmission between the routing paths becomes impossible.

To compensate for this, in the past, special buses were added between the switch elements in each call step, and the links between the call route sections were shared by multiple routing routes, thereby increasing the number of detours within the routing route. Methods have been proposed to deal with failures that occur in
Eng, P, Yew, C, Zhu.

“A Fault-Tolerant S c
heme for Multistage Int
erconnection networks.

12th International S
ymposium on Computer Arc
Hitecture, June (1985).

reference).

FIG. 11 is a circuit diagram showing an example of the configuration of a self-routing communication path (number of incoming lines=number of outgoing lines=8) shown in Reference Document 2.

In communication path stages 1 and 2 in the same figure, a special bus for communication between switch elements is added to each of them. The special bus on call route 1 is 141-0-141-3.
For example, when link 122-0 fails,
Communication information requesting connection to the next stage of communication path from switch element 111-0 via link 122-0 is transmitted to switch element 11 via special bus 141-0.
1-1, and is transferred from this switch element to the next stage. Similarly, when the link 122-4 fails, the communication information is connected to the switch element 111-1 in the same stage via the special bus 141-0.

Furthermore, when the output link desired by the communication information cannot be used in the switch element 111-1, 14
It is connected to the next switch element in the same stage, 111-2, through the special link 1-1.

[Problem to be solved by the invention]

In the conventional self-routing communication path as described above, when a failure occurs in the routing path, in order to avoid the failure location, communication information is routed immediately before and at the same time as the failure occurred.
Since the communication information is distributed to the switch elements within the stage, link conflict occurs between the communication information within the switch elements to which the communication information is distributed, reducing the possibility that the communication information will reach the outgoing line. As a result, the outgoing line usage rate of the switch circuit network is extremely low.

In order to solve the above-mentioned problems, an object of the present invention is to provide a self-routing communication path that can set a detour in a manner that minimizes the reduction in the output line usage rate of the switch network even if a failure occurs in the routing path. Our goal is to provide the following.

[Means to solve the problem]

In order to eliminate the above-mentioned drawbacks of the self-routing channel, the present invention adds a new technical means, a plurality of sub-switches, in each channel stage. And that is its most important feature.

[Effect]

In the present invention, by adding a plurality of sub-switches in each call step, even if a failure occurs in the routing path, communication information is passed through the sub-switch to the next switch without being distributed to other switch elements in the same step. Since the communication information can be transmitted to the switch elements in each stage, link conflicts between communication information within the switch elements are reduced, and a decrease in the outgoing line usage rate of the switch circuit network can be prevented.

At the same time, the number of routing paths connecting any incoming line and any outgoing line is greatly increased by adding sub-switches, so fault tolerance is significantly improved. In addition, since the subswitch effectively works to avoid link conflicts between communication information, even in normal conditions when there is no failure in the self-routing communication path, the outgoing line usage rate of the switch circuit network is lower than before the addition of the subswitch. Improved.

〔Example〕

Next, the present invention will be explained in detail based on the drawings.

FIG. 1 shows an embodiment of the present invention (number of incoming lines=number of outgoing lines=8).
) is a circuit diagram showing.

In the figure, 230-0 to 230-7 are incoming lines, 231
-0 to 231-7 are outgoing lines, 200-0 to 200-7 are connection information adding circuits, and 201-0 to 201-7 are connection information removal circuits.

The basic circuit (switch element) that determines the link to be connected based on the connection destination address given in the communication information is displayed as -0-a (k=1~3.a-　~3) in 21. There are switch elements and other sub-switches, and the switch element indicated by -1-a in 21 is called a rank r sub-switch, and the switch element indicated by -2-a in 21 is called a rank 2 sub-switch.

Here, switch element 21 -0-0~21-
2-a constitutes communication path stage k, and generally stage k is provided with sub-switches up to rank (, n-k). 22
1-0-0~224-0-7° and 240-0-0~2
41-0-3 is a link that connects the connection information adding/removing circuit and the switch element, all of which are numbered sequentially from the top.

Binary notation is used to represent the locations where links are connected, numbering according to the following rules: The position of the link connecting to the basic switch element is
As shown in FIG. 2, except for the links that communicate with the sub-switch (the output to the sub-switch is especially called a detour output terminal), the links are 0, 1, etc. from the top. ......, N-
Numbered 1, this is expressed as a binary number (pn-+l
-z l ..., Po) are shown in Table n, and the detour output terminals are (pn-z) in order from the top.

p□3. ..., po), and the input terminals from the sub-switches are O(P-z, T'fi-3, ..., in order from the top).
po).

For rank r subswitches, the positions of links connecting to subswitches are r (p, l−
+-r, pn-2-r+..., Po),
The detour output terminals are arranged in order from the top (Pn-Z-r +
Pn-1-r +..., po)', crank r+1)
The input terminals from the sub switch are r (pn
-z-1+ P n-3-r r..., po).

Here, one example of a link wiring method between switch elements will be explained using FIG. 2.

In general, the output terminals (P,, -+, Pn-z,...
・p6) is the input terminal (pn-+) of the communication line stage (k+1)
.

..., P7-□r, po, Pn-e, ..., p,)
is connected to the detour output terminal (pn-z + Pf
i-3,...po) are the input terminals 1-(+)n-z, 'P--:l,..., of the rank r sub-switch.
Po), the output terminal r- of the rank r sub-switch
(p・-tor・Pn-2-r+"'po) is the input terminal (r-1) (Po, Pn-
+-r,...p+), the detour output terminal rC
pn-z i+P n-ff-r +...po) is the input terminal (r+1)(
Pn-2-r + Pn-3-r + ...po)
connected to.

However, in the case of communication line stage (n-1), the output terminal 1 of the rank l sub switch (P, , -z, pn-: +,
...po) is the input terminal 0 (pn-t
+ Pn-3+..., Po).

Although one example of the link wiring method has been shown above, the wiring between basic switch elements is not limited to this example.

For example, the effectiveness of the present invention is not impaired in any way even if the links between basic switch elements are replaced with modified data manipulators (see Reference 2 above). Similarly, the wiring of the output link from the sub-switch is not determined arbitrarily;
The figure shows only one example.

Next, the operation of the switch element will be explained.

FIG. 3 is an enlarged view of a part of FIG. 1. here,
The communication information input from the incoming line of 221-0-0 is
The connection to the outgoing line -0-2 will be explained as an example.

The position of link 223-0-2 is expressed as (01) in FIG. 3, so the connection destination address with communication information is (01).
). Switch element 211-0-0 reads O in the first bit of the connection destination address and attempts to set a path to 222-0-0, which is the upper output link of the switch element, but at this time, the following two points are required. Confirm and then connect.

The first point is whether the link 222-0-0 is already being used by other communication information, which can be easily determined by referring to the path setting information in the switch element. The second fish is link 222-0-0
and whether the connection modules at both ends of the link can be used. To check this, switch element 211-0-0 connects switch element 212-0-
0 through link 222-0-0. When switch element 212-0-0 receives a connection request signal, it immediately returns a connection authorization signal using the same link. The switch element 211-0-0 determines that the link 2220-0 and the connection modules at both ends of the link are usable only when the connection authorization signal is received.

After the above two points are confirmed, switch element 21
1-0-0 is link 221-0-0 and link 222-0
-0 and send the connection information to the switch element 212-
Send to 0-0. At this time, by discarding 0 in the first bit of the connection destination address and transmitting from the second bit, connection control can be performed based on only the first bit in the subsequent communication path stages.

Alternatively, there is a method in which the first bit of the connection destination address is not discarded, but is cyclically shifted and then sent. FIG. 4 shows how the connection destination address changes at each communication channel stage.

In Figure 4, the connection destination address (101) is used as an example.
) is a cyclic shift method.

Returning to FIG. 3, if a connection approval signal is not obtained within a certain period of time after sending a connection request signal, the switch element 12
1100 determines that the link 222-0-0 or the connection modules at both ends of the link are unusable, and the link 211
-0-0 is a detour to the subswitch link 221
Connect to -1-0. At this time, the destination address is sent as is without any processing.

As shown in FIG. 3, if the link 222-0-0 is out of order, the communication information is transferred to the rank r subswitch 211-1-
0, and this subswitch reads 0 in the first bit of the destination address, confirms the above two points, and then sends the second bit and subsequent bits of the destination address to the link 240-0-0.

This link is connected to switch element 212-0-0, and the same operation is performed in switch element 212-0-0 to connect communication information to link 223-O-2.

After the above steps, the communication information is transferred to the target outgoing line 223-0.
-2 is reached.

Although the failure of link 222-0-0 is taken up as an example in Fig. 3, according to the above principle, not only the link failure but also the failure of the connection modules at both ends of the link, and the unit of the switch element including the connection module can occur. It is also possible to deal with breakdowns and set up detours.

FIG. 5 is a diagram showing an example of the configuration of a switch element.

The switch element may have any configuration as long as it can realize the operation according to the above principle, and in fact, a plurality of configuration examples can be considered, and FIG. 5 will be described as one example.

The switch element is composed of a human control circuit (IC), an output control circuit (OC), a central control circuit (CC), and a connection control circuit (CK). When the input control circuit (IC) receives the communication information, it sends the first bit of the connection destination address to the central control circuit (CC).

The central control circuit (CC) determines whether the output link corresponding to this decision bit is used.
) to determine the path setting information. If it is already in use, the input control circuit (I
C) is instructed to output to the subswitch. If it is in an unused state, a test request signal is sent to the connection control circuit (CK) to see if it is possible to connect to the next stage using the desired output link. The connection control circuit (CK) sends a connection request signal to the next stage switch element through the output control circuit (QC) of the output link, makes a decision within a certain period of time, and returns the result to the central control circuit (CC). do. If available, the central control circuit (CC) directs the input control circuit (IC) to connect to this output link. If unavailable, the central control circuit (CC) instructs the input control circuit (IC) to output to the subswitches. When the central control circuit (CC) instructs the input control circuit (rc) to establish some kind of connection link, the central control circuit (CC) holds that information as bus setting information in any case. Assuming that communication information is input to a plurality of input control circuits (ICs) at the same timing, for example, the input control circuits (ICs) are assigned priorities for processing.

Furthermore, if a link conflict occurs in the output to the subswitch, retransmission is instructed for communication information with a lower priority. Alternatively, it is also conceivable to provide a buffer in the output control circuit (OC) connected to the output link to the subswitch. In this way, the switch element is configured not only to detect and avoid failures (but also to effectively deal with link conflicts between communication information using detours).

Here, a description will be given of the routing path when a failure also occurs in the sub-switch. FIG. 6 is a diagram showing a part of FIG. 1 extracted. When communication information with connection destination address O is input from link 221-0-0, switch element 211-0-O attempts to connect to link 222-0-0, but when this link is out of order. is sent to link 221-1-0, which is a detour. At this time, the connection destination address is not processed, so the first bit is O.
It remains as it is.

Rank 1 subswitch 211-1-O that received communication information
links communication information based on the first bit 0 240-0
-0, but if that link is out of order, it is sent to link 221-2-0, which is a detour. At this time, the destination address is not processed, so the first bit remains O. Connected rank 2 sub switch 2
At 11-2-0, the first bit ()O is determined and the communication information is connected to link 240-1-0.

What should be noted here is that connection control is performed in every switch element based only on the 0 of the first bit of the connection destination address.

In other words, regardless of the number of times detours are used within the same stage of the communication path, the minimum unit of information required for connection control is only one bit. This provides great advantages such as simplifying the connection procedure and simplifying the connection information.

FIG. 7 is a diagram showing the overall operation of the self-routing channel. In Figure 7, the destination addresses are (001) and (100).
), (111) are shown. The operation of each switch element is as described above. The example (001) shows a routing path when a failure occurs in a link on the way. In the examples (100) and (111), information is manually collected from a position where link conflict occurs on communication path 1,
We also showed how subswitches work effectively to avoid link conflicts between communication information.

As can be seen from the specific example in FIG. 7, in the present invention, by adding a plurality of sub-switches within each call step, even if a failure occurs in the routing path, communication information can be transferred within the same step. The signal can be transmitted to the next stage switch element through the sub-switch without being distributed to other switch elements. Therefore, the possibility of link conflict between communication information occurring within the switch element is significantly reduced compared to the self-routing channel of Reference 2.

Furthermore, even if a link conflict occurs between pieces of communication information, it is possible to avoid the link conflict due to the structure of the switch element, so the outgoing line usage rate of the switch circuit network is extremely high.

Figure 8 compares the outgoing line usage rates before and after introducing the sub-switch. It can be seen that the difference between the two widens as the switch size increases. Furthermore, by adding sub-switches, the number of routing paths connecting any incoming line and any outgoing line is greatly increased, so fault tolerance is increased (improved).

FIG. 9 is a diagram showing a method of adding self-routing communication paths according to the present invention. The 8x8 switch network shown in Figure 1 is 16
This is an expanded version of X16. Two 8.times.8 switch networks 700 and 701 are arranged in parallel, a communication path stage for expansion is placed on the input side of the switch networks, and the switch elements are connected according to the link wiring method described above. (It is necessary to prepare a third input terminal for the detour connection on the input side switch element of the 8x8 switch network.) Generally, when expanding the NXN switch network to a 2NX2N scale, It is only necessary to arrange two NXN switch networks in parallel, place a 2N manual expansion channel on the input side of the switch network, and connect the switch elements according to the link wiring method described above. At this time, if an input terminal for connecting a detour is prepared on the input side of the switch element of the expansion channel stage, further expansion becomes possible. In this manner, the self-routing communication path of the present invention can be easily expanded without replacing existing link wiring.

The embodiments described above are merely illustrative of the invention. Other devices may be devised by those known within the principles of the present invention. for example,
A switch network with a different link wiring algorithm but with additional sub-switches similar to the present invention (for example, a link wiring with the modified data manipulator and an additional sub-switch), and a switch where the position of the sub-switch is simply replaced. One in which the wiring of the output link from the sub-switch has been changed to distribute the load, one in which the function of the central control circuit (selection of communication information) is distributed among the output control circuits, and one in which the connection control circuit is Examples include those that simply change the control.

〔Effect of the invention〕

As explained above, according to the present invention, by adding a plurality of sub-switches in each call step, even if a failure occurs in the routing path, communication information is transferred to other switch elements in the same step. Since the information can be transmitted to the next switch element without being distributed to other switch elements, link contention of communication information within the switch element is reduced.
This has the advantage of keeping the outgoing line usage rate of the switch network at a high level.

At the same time, the addition of sub-switches greatly increases the number of routing paths connecting any incoming line and any outgoing line, which has the effect of significantly improving fault tolerance.

In addition, since the subswitch effectively works to avoid link conflicts between communication information, even in normal conditions when there is no failure in the self-routing communication path, the outgoing line usage rate of the switch circuit network is lower than before the addition of the subswitch. Improved. Furthermore, as can be seen from the fact that the number of judgment bits in each stage is the minimum number of 1 bit, it can be operated with a very simple algorithm and has excellent expandability, so its range of application is wide.

[Brief explanation of drawings]

FIG. 1 is a circuit diagram showing an embodiment of the present invention, FIG. 2 is an explanatory diagram explaining the link wiring method, FIG. 3 is a circuit diagram showing the route through which communication information passes through the switch element, and FIG. 1st
An explanatory diagram showing changes in the communication information transferred through the switch circuit network shown in the figure, FIG. 5 is a configuration diagram showing a specific example of the configuration of the switch element in FIG. 1, and FIG. FIG. 7 is a circuit diagram showing the route through which the communication information in FIG. 4 passes through the switch circuit network. FIG. FIG. 9 is an explanatory diagram showing a method for adding a self-routing channel according to the present invention, and FIGS. 10 and 11 are diagrams showing conventional examples of self-routing channels. FIG. Explanation of codes 230-0 to 230-7...Incoming line, 231-0 to 2
31-7... Outgoing line, 200-0 to 200-7...
Connection information adding circuit, 201-0 to 201-7... Connection information removal circuit, 21 k-0-a (k=1 to 3°a=
O~3)...Basic switch element, -2- to 21
a...Rank 2 sub switch

Claims (1)

[Claims] 1) N (=2^n, where n is a natural number) incoming lines and N outgoing lines via a plurality of communication paths and links connecting them. In a self-routing communication path that connects and transfers communication information from an arbitrary incoming line to an arbitrary outgoing line according to connection information attached to the communication information, each communication path stage has the input communication information. It is composed of a plurality of switch elements that determine the next link to which input communication information should be connected based on the judgment bit in the connection information, and the plurality of switch elements that constitute each channel stage are 0-order, 1-order, Next,..., m (however, m≧1)
It consists of the following switch elements, and has one or more output terminals for next-stage connection corresponding to the judgment bit in the connection information, and except for the switch element with the highest order in each channel stage, the output terminal for the next-stage connection corresponds to the judgment bit in the connection information. In addition to the output terminal, it has a detour output terminal for setting a detour, and each switch element is configured to transfer communication information to the next one due to a failure of a link or the next switch element, or a link conflict between communication information. If it is determined that the connection cannot be made to the next stage, the communication information is outputted to the switch element whose order is one higher than that through the detour output terminal, and the switch element that receives the input of the communication information newly connects to the next stage. 1. A self-routing communication path, characterized in that a routing path to a communication path is set. 2) A self-routing channel according to claim 1,
The number of the communication path stages is n, the highest order m of the switch element in the k-th communication path among them is given by (n-k) or less, and the 0th order switch element is the basic switch element, and the 1st order to m When the next switch element is referred to as a rank 1 to m sub-switch, the basic switch element is provided so that a path from any incoming line to any outgoing line can be set by itself, and each communication path stage is , N/2^i (however, i≧
1) elementary switch elements, rank 1 sub-switches include N/(2^i^+^
1) or less, generally rank r subswitches (however, 1≦
r≦m) is N/(2^i^+^) at each communication path stage.
r) a self-routing channel comprising: 3) A self-routing channel according to claim 2,
The k-th channel stage k is the same as the preceding channel stage (k-
a set of switch elements that receive communication information output from 1) corresponding to 0 in the connection information bit;
a set of switch elements that receive the output, and one of the output terminals for connecting to the next stage of the communication path stage (k-1) is connected to the set of switch elements of the communication path stage k that receives the bit 0 output; A self-routing channel, characterized in that one is connected to a set of switch elements of channel stage k which receives a bit 1 output. 4) A self-routing channel according to claim 2,
The switch element performs connection control based on the first bit of the connection information included in the communication information, and processes the first bit of the connection information when sending the communication information to the next stage of the communication path. Self-routing characterized by not processing the connection information when sending it to the switch, making it possible to control the connection using the same connection information and only one bit within the same communication path unless it is sent to the next stage. Call path.