JPH042238A - Self-routing network by input buffer system - Google Patents

Abstract

PURPOSE:To reduce the aborting rate of a data cell to be transmitted by providing a data distributing circuit connecting each input terminal and each input buffer selectively properly and distributing a data cell inputted to each input terminal to each input buffer to the pre-stage of the input buffer. CONSTITUTION:Each of input ports #1-#4 is connected to each input of a data distributing circuit 11 and each data of the data distributing circuit 11 is connected to each input of each of input buffers B1-B4. Moreover, each output of each input buffer B is connected to each input of a switching network S. Then the transmission path for a data cell inputted to each of the input ports #1-#4 is selected by the data distributing circuit 11 and fetched alternately to each of the input buffers B1-B4. Thus, even when a retransmission requirement issues frequently to a specific input buffer, the data cell inputted newly is distributed to each input buffer and fetched by the data distributing circuit. Thus, the aborting rate of the data cell is reduced and the transmission performance is improved.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a self-routing network using an input buffer method in which input data taken in through an input buffer is output to a predetermined output terminal by self-routing. .

[Conventional technology]

Conventionally, as a self-routing network using this type of input buffer method, for example, the one shown in the document "Prototype production of ATM switch using input buffer method J (1988 Institute of Electronics, Information and Communication Engineers Autumn National Conference, B-193) be.

The self-routing network in the same document is summarized in Section 5.
As shown in the figure. Input port #1. The data cells input to #2° to #n are input to input buffers Bl, B2 . ~ Once taken into Bn. The hatched areas of input buffer B in the figure indicate data cells that have been taken in and stored, and the blank areas indicate empty areas in which no data cells have been stored yet. The captured data cells are sent to the switch network S at each cell sending timing. Here, when a cell collision occurs within the switch network S, a cell retransmission request signal is sent to the input buffer B.
The input buffer B retransmits the stored data cells to the switch network S at the next cell transmission timing.

[Problem to be solved by the invention]

In each of the above conventional input buffers B, input data cells are transmitted as indicated by the arrows and stored in the input buffer B specified by the input port. Therefore, new data cells are input one after another to the input buffer Bi shown in FIG. 6, where cell retransmission processing is frequently performed.
May accumulate. As a result, even though there is an input buffer Bj with a blank space, the accumulated data cells exceed the storage capacity of the input buffer Bi, and when a new data cell is input, the input buffer Bi overflows. You'll end up leaving. Therefore, data cells to be transmitted are lost, resulting in a high cell discard rate.

[Means for Solving the Problems] The present invention has been made to solve the above problems, and it connects each input terminal and each input buffer as appropriate, and connects the data input to each input terminal. A data distribution circuit for distributing cells to each input buffer is provided before the input buffer.

[Effect]

Even if there are frequent retransmission requests to a particular input buffer, newly input data cells are distributed to each input buffer and taken in by the data distribution circuit.

〔Example〕

FIG. 1 shows a self-routing network using an input buffer method according to an embodiment of the present invention.

Each input port #1 to #4 is connected to each input of a data distribution circuit 11, and each output of this data distribution circuit 11 is connected to each input of each input buffer 81 to B4.

Further, each output of each input buffer B is connected to each input of a switch network S, and each output of the switch network S is connected to each output port.

The switch network S has a plurality of 2×2 unit switches with 2 inputs and 2 outputs as one unit of its configuration, and has m columns and n rows (m
, n is a natural number), thereby forming a Banyan network. Each 2×2 unit switch determines its own connection state according to the destination information stored in the data cell. The input data cells are transmitted to each output port and outputted by each unit switch under its own routine. In addition, a similar unit switch is connected in the opposite direction to the cell transmission direction, and when a data cell collision occurs in a certain unit switch, this collision information is transferred to the input buffer B that sent the cell. will be returned to.

At this time, the collision information is sent back by each unit switch provided in the opposite direction, following the same route as that through which the cell was transmitted.

Note that this collision occurs because each data cell tries to proceed in the same direction of travel according to each destination information input to each unit switch.

Each of the input buffers B1 to B4 temporarily stores and accumulates data cells input to each input port, and sends out the accumulated data cells to the switch network S in synchronization with the cell sending timing. Furthermore, when each input buffer B receives the above-mentioned collision information, it retransmits the same cell that was sent out at the previous cell sending timing again at the next cell sending timing.

The data distribution circuit 11 is a circuit that appropriately switches and connects each input port #1 to #4 and each input buffer 81 to B4, and the switching and connection is performed as follows. In other words, the switching connection state of this data distribution circuit 11 is as follows.
It sequentially switches to the four states shown in FIG. 2 in synchronization with the cell sending timing, and does not remain stationary in one state. That is, the connection state changes at the time interval when one data cell is sent out, and the state of the data distribution circuit 11 changes for each data cell.

The connection state shown in Figure (a) is input ports #1 to #
4 is connected straight to input buffers B1 to B4. In other words, a data cell input to input port #1 is transmitted as is to input buffer B1, and a data cell input to input port #2 is transmitted as is to input buffer B2. In the connection state shown in the same figure (b), input port #1 is connected to input buffer B2.
and input port #2 is connected to input buffer yB3. The connection state shown in FIG. 2C is such that input port #1 is connected to input buffer B3, and input port #2 is connected to input buffer B4. The connection state shown in FIG. 4D is such that input port #1 is connected to input buffer B4, and input port #2 is connected to input buffer B1.

In such a configuration, the data cell input to each input port #1 to #4 has its transmission path switched by the data distribution circuit 11, and is transferred to each input buffer B1 to B4.
are taken in alternately. For example, a data cell input to input port #1 is first taken into input buffer B1, then taken into input buffer B2 at the next cell sending timing, and the input buffer B to be taken in is switched sequentially, and all input buffers are taken in. It is taken into B. Data cells input to other input ports #2 to #4 are also sequentially fetched into all input buffers B in the same manner.

The data cell taken into the input buffer B is sent to the switch network S, its transmission path is switched under self-routing, and the data cell is output to a predetermined output port according to the destination information stored in the cell.

Here, if a cell collision occurs in a certain unit switch in the switch network S, the collision information is returned to the input buffer B that sent the cell;
Even if collision information is sent back frequently, cells will overflow less often in this input buffer B than in the past. This is because new data cells taken in at the next cell sending timing are distributed to all input buffers B as described above. In other words, newly input data cells are not intensively fetched into the input buffer B that is executing the retransmission process, but are distributed and fetched into other input buffers that have free space.

As a result of simulating a self-routing network using the input buffer method to which this embodiment is applied, the following good results were obtained, and the effectiveness of this embodiment was confirmed.

The self-routing network in this simulation is 1
It has 6 input terminals and 16 output terminals, and its data distribution circuit is shown in FIG. That is, data distribution circuit 1
2 is composed of logic circuits sl, s2° to s16, and each input of each logic circuit S has an input port #1. #2. ~#16
is connected. Further, the output of the counter 13 is connected to each input of each logic circuit S.
3, a clock pulse CLK is input.

Further, each output of the logic circuit S is connected to each input buffer Bl.

B2. ~B16 are given to each input. The depth (memory capacity jl) of each of these input buffers 7B is 20 data cells, and up to 20 data cells can be stored.

The counter 13 detects the cell sending timing by counting the clock pulse CLK, and outputs a timing signal to each logic circuit S when the sending timing comes.

When each logic circuit S receives this signal, it switches the connection state between each input port and each input buffer in the same manner as in the above embodiment.

As a result, each input port #1. #2. ~#16 i,
: The input data cells are in each input buffer Bl.

B2. .about.B16 and are taken in, and even if there are frequent retransmission requests to a specific input buffer B, overflow of data cells is reduced.

Table 1 below shows the results of comparing and measuring the data cell discard rate by setting the input port occupancy to various values between the circuit based on this simulation and the circuit with a conventional configuration that does not have a data distribution circuit. The result was that Here, the occupancy rate refers to the ratio of the time when data cells enter the input port and the time when data cells do not enter, and the occupancy rate is 1.00.
This means that the data cells enter the input port one after the other without any interruption, and the occupancy rate of 0.00 means that the data cell does not enter all the heat into the input port. . Furthermore, the discard rate is the proportion of data cells that overflow from the input buffer B and are lost, and the smaller the value, the better the transmission performance of the self-routing network is.

Table 1 Since the input port occupancy rate in actual use is low, as can be understood from the above table, it is confirmed that the simulation circuit which is an application of this embodiment is more effective.

In the description of the above embodiment, a self-routing network with four inputs and four outputs has been described, but this is for convenience of explanation and is not necessarily limited to this. For example, it may be a self-routing network with n inputs and n outputs (n is a natural number) shown in FIG.
By configuring the input buffers Bl and 82 to Bn in accordance with the embodiments described above, the same effects as in the embodiments described above can be obtained.

〔Effect of the invention〕

As described above, according to the present invention, even if retransmission requests are frequently made to a specific input buffer, newly input data cells are distributed to each input buffer and taken in by the data distribution circuit.

For this reason, new data cells are less likely to overflow from the input buffer as in the past, the data cell discard rate is reduced, and a self-routing network with good transmission performance can be provided.

[Brief explanation of drawings]

FIG. 1 is a configuration diagram showing a self-routing network using an input buffer method according to an embodiment of the present invention, FIG. 2 is a diagram showing the connection state of the data distribution circuit shown in FIG. 1, and FIG. FIG. 4 is a circuit configuration diagram showing a data distribution circuit in simulation based on the embodiment; FIG. 4 is a configuration diagram showing another example according to the present invention; FIG. 5 is a configuration diagram showing a self-routing network using a conventional input buffer method; FIG. 6 is a configuration diagram showing a self-routing network to explain the problem. 11...Data distribution circuit, #1-#4...Input port, B1-B4...Input buffer, S...Switch network. Structure of Example Figure 1 Representative Patent Attorney Yoshi Hasekima
Salt 1) Tatsuya Data oscillation circuit connection state Figure 2 Degu minutes circulation diagram Figure 3 Other items 9 Conventional configuration Figure 5 Explanation of the task C Ken

Claims (1)

[Claims] Data cells input to a plurality of input terminals are taken into a switch network via a plurality of input buffers, and the data cells are outputted to a plurality of locations based on destination information stored in the data cells. In a self-routing network using an input buffer method for outputting to a terminal, a data distribution method is provided in which data cells input to each input terminal are distributed to each input buffer by appropriately switching and connecting each input terminal and each input buffer. A self-routing network using an input buffer method, characterized in that a branch circuit is provided at a stage before the input buffer.