JPH0276436A - Multiple address packet switch - Google Patents

Abstract

PURPOSE:To prevent the generation of collision between packets and the generation of internal congestion by connecting a memory for generating an output to each output line in each output line. CONSTITUTION:Packets inputted from input lines 101 to 103 are multiplexed by a time division multiplex part 104. The time-dividedly multiplexed packets are outputted to a bus 105 one by one and the identification(ID) numbers of the packets are read out. A bit map indicating the output lines to which respective packets are outputted in accordance with respective ID numbers is stored in a table 107. Bit map information is transmitted to gates 111 to 113 through control lines 108 to 110, outputted only to a required output line and stored in memories 121 to 123. Consequently, the generation of collision between excess packets in the prestage of a switch can be prevented and the generation of congestion in a high load can be also prevented.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a technique for configuring an efficient broadcast packet switch.

(Prior art) As a paper on broadcast control in conventional high-speed packet switches, ■"Design" published by J. S. Turner at IEEE Infocom '86 in 1986.
of a Broadcast Patch Ne
work”■19th April IEEE Infocom
In '88, Y, Eng and others announced "M
ulticast and broadcast 5e
rvice in aKnockout 5w1tch
” ■July 1988 Institute of Electronics, Information and Communication Engineers Exchange Study Group 5SE8
A proposal for the rATM switching architecture announced by Endo et al. on 8-56. ■ ``A proposal for the rATM switching architecture announced by Endo et al.
5DN Teleconferencing System] is a typical example.

Among them, document (2) copies broadcast packets at the front stage of a packet switch. In other words, in the case of 1:M broadcasting, one packet is duplicated M times at the stage before the switch (in this paper, copy network J), and is input in parallel to arbitrary M input lines of the switch. However, in this method, one packet is copied multiple times before entering the switch, so of course the load on the switch (referred to as the "routing network" in this paper) is as follows: It becomes M times.

If this M times the load is applied to the entire switch, it may cause collisions with other broadcast packets within the switch, and may affect not only the outgoing lines that should be broadcast, but also all the buffers that the broadcast packets pass through within the switch. The switch becomes very congested due to the added load.

Originally, the switch in Document (2) tends to cause congestion when the load is high, so a problem arises in that it is difficult to accommodate a large number of broadcast calls with a large number M of broadcasts. This is essentially caused by the fact that packets are copied before the switch.

In response to this, documents ■, ■, and ■ propose a method in which packets are copied inside the switch.

Documents (1) and (2) propose a method in which packets from multiple incoming lines are first condensed and stored in memory, and then the packets are copied. In Document 2, the line concentrator is called a multicast module, and broadcast packets are stored in a shared output buffer within this module. In Document (2), time-division multiplexing is performed once, and broadcast packets are stored in a broadcast packet queue in a shared memory. Then, when outputting from these memories, the packets are broadcast to multiple outgoing lines. Document (2) uses a broadband bus, and Document (2) uses a time division multiplexing/demultiplexing device. The method of this document (■■) will be explained in more detail with reference to FIG. Document ■■ mainly deals with packet communication with 1:1 connection, and also has a 1:M broadcast packet switch function added. FIG. 2 shows only the added broadcast packet switch unit among them, and shows the principle of the broadcast system of Document (2). The difference from document ■ is that line concentration is performed by a concentrator instead of time division multiplexing, and the table for the broadcast filter on the outgoing line is
Although they are arranged in a distributed manner on each outgoing line, the operation is basically the same as that shown in this figure in that they are once stored in the memory and then broadcast. Packets input through input lines 201 to 203 are time-division multiplexed by time-division multiplexing section 204 and stored in memory 205 .

Broadcast control is performed when reading from memory.

The packet is copied to all outgoing lines by branching device 206. At that time, filtering control is performed so that the signal is output only to the desired outgoing line. This is achieved by table 208. Each packet has a call identification number.

The identification number is read through the signal line 207. The table 208 holds information in the form of a bitmap about which outgoing line a packet should be output to or not for each call identification number. For example, 1 for the outgoing line (1, 2, ..., N),
0. ..., 1), 1 corresponds to output, 0 corresponds to no output, and means that output is not output to output line 2, but output to output line 1 and N. The outgoing line to which this broadcast should be broadcast is known from the identification number read in step 207. This result is notified to gates 212-214 via control lines 209-211. The gates 212-214 output the packets copied to all outgoing lines 215-217 in 206 only to desired outgoing lines. Now, the output speed of packets from this memory 205 is now equal to the speed V of the outgoing line. The time it takes for one packet to be read from the memory at a speed V is defined as a unit time. Here, let M be the average number of broadcast calls. In other words, one packet input from the incoming line is output as M packets on the outgoing line. Here, if the line utilization rate of broadcast traffic on the outgoing line is pB and the number of outgoing lines is N, an average of pBXNIM packets are written to the memory 205 per unit time. However, since only one packet is read from the memory per unit time, pBXN×-≦1 M must hold true. In other words, pB≦−, which limits the throughput of broadcast traffic. Intuitively, one packet from memory 205 is M
This is because a situation occurs in which the packet is not output to the remaining NM outgoing lines while the packet is being copied to the remaining NM outgoing lines. The essential cause of this phenomenon is that the lines are once concentrated in the memory 205 and then the packets are attempted to be broadcast. Therefore, compared to 1=1 communication traffic, if the broadcast traffic is small (pB is small) or if the number of broadcasts is large (M
The problem is that this switch can only be used in cases where the Now, in Document 2, the table 208 is installed separately for each outgoing line (first address filter in the document), and a buffer is further installed after the outgoing line 215, but essentially the above problem remains. There is. The method described in Document ■ does not have any restrictions on congestion within the switch as in Document ■ or throughput as in Document ■■, but it does have a problem with the amount of hardware. This method is shown in FIG. Packets input from incoming lines 301 to 303 are stored in table 321 for each incoming line.
~323 is accessed. Tables 321 to 323 have the same function as table 208 in FIG.
Output from 27 to 329. Based on this bitmap, each packet is sent to the gate 34 corresponding to each outgoing line 311 to 313.
1 to 349 and stored in memories 351 to 359. For example, a packet input from the incoming line 301 is sent to the outgoing lines 311 and 313 according to (10...1) in the table 321.
The gates 341 and 343 are opened so that the data is output to the memory 351 and 353 at the same time. Each outgoing line polls the memory for each incoming line and outputs a packet. For example, outgoing line 311 searches memory 351.354.357. Although this method is easy to control as described above, it has the same problems as the inverse method. Table 32 where the number of incoming and outgoing lines is N
Each of 1 to 323 has a bit width of N, and the number of tables required is N. Therefore, the memory capacity constituting the table increases in proportion to N2.

Also, like the number of memories 351 to 359, it is proportional to N2.

In other words, the amount of hardware is extremely large. Next, since the tables 321 to 323 are distributed, table management for broadcasting becomes complicated for the CPU 332.

The CPU needs to manage N tables using the bus 330, and updating the tables takes time.

(Problems to be Solved by the Invention) As described above, the broadcast packet switch according to the prior art is: 1) Internal congestion occurs. 2) There is a limit on the throughput of broadcast traffic. 3) The hardware scale is large and table management is difficult. Hateful.

It has the following problems. The present invention provides a method for realizing a broadcast packet switch in an efficient manner, with small hardware scale, easy table management, and a simple method that does not cause any of the above-mentioned problems.

(Means for solving the problem) A broadcast packet switch that has multiple incoming lines and multiple outgoing lines and outputs a packet from the two incoming lines to any one of the outgoing lines based on the identification information of the packet. can be,
The packet switch includes means for time-division multiplexing packet signals and identification information of the packets from a plurality of incoming lines, means for broadcasting the time-division multiplexed packet signals to all outgoing lines, and means for broadcasting the time-division multiplexed packet signals to all outgoing lines. a first memory means for storing the packet signal and outputting it to the plurality of outgoing lines for each outgoing line; a second memory means for holding information as to whether or not the packet should be output; and from the time-division multiplexed identification information, each of the plurality of outgoing lines held in the second memory is configured to output the identification information; The method further comprises means for reading the information indicating whether or not the packet containing information should be output, and means for controlling whether or not the packet is stored in the first memory means according to the read information. By configuring a broadcast packet switch characterized by the following, it is possible to provide a packet switch that is simple to control and can efficiently perform broadcasts.

(Function) In the present invention, since broadcasting is performed inside the switch, the load on the switch is purely applied to the outgoing line to which broadcast packets are output. Therefore, switch congestion does not occur, and broadcast packets do not affect other packets.

Further, in the present invention, input packets are time-division multiplexed, then broadcast control is performed, and buffering is performed in memory on the outgoing line to be output. In other words, the broadcast packets are not stored in the memory in a concentrated manner, but are distributed for each outgoing line and input into the memory.

Therefore, unlike the conventional example, there is no limit to the amount of broadcast traffic pB or the number of broadcasts M on the outgoing line. In other words, it can also be used as a broadcast-only switch. Also, since the control table is time-divided by all incoming lines, it can be concentrated on one. Therefore, the amount of hardware required for the table is small, and management by the CPU can be integrated. The access speed of the broadcast table is naturally
It is N times more than . However, if the packet length is L, the incoming line speed V is sufficient to perform at a speed of VXN, so this does not become a bottleneck in the speed of the broadcast table.

(Example) An example of the present invention is shown in FIG. Packets input from incoming lines 101 to 103 are multiplexed by time division multiplexing section 104. Here, if the input line speed is V, the speed of the time division multiplexing section is NX
V (number of input lines; N). An example of the format of a packet input in this manner is shown in FIG. Packet 4
0 indicates VCI (Virtual Channel) in the header.
Packet identification number 4 called Identifier
2 and a data section 41. The time-division multiplexed packets are broadcast to all outgoing lines 131 to 133 via bus 105. Here, broadcast filtering control is performed as follows. As the time-division multiplexed packets are output one by one to the bus 105, the identification number (VCL) of each packet is read (106). The table 107 holds a bit map indicating to which outgoing line the packet should be output for each identification number. According to the figure (10...1
), so it is not output to outgoing line 132, but outgoing lines 131 and 133.
This means that it will be output to. This bitmap information is transmitted to gates 111-113 via control lines 108-110, and packets are output only to desired outgoing lines. The output packets are stored in memories 121-123. The writing speed to the memories 121 to 124 is NXV. On the other hand, the voltage read to the output lines 131 to 133 is V. This switch has a memory for each of the N outgoing lines, so data is written to the memory at the same time. Only one table is required and table 107 can be easily managed by CPU 140. The memory and table are accessed by packets arriving at NY speed, but if the packet length is L, it is easy to reduce the speed to NXV speed by parallel expansion.

(Effects) This switch has a memory for each output line and broadcasts immediately before it. Therefore, there is no collision between extra packets at the front stage of the switch, and no collision occurs under high load.

Also, since the speed of the multiplexing section is NXV, each memory is provided at each outgoing line, and the write speed of the memory is also NXV, the broadcast traffic throughput limit is 10 regardless of the number of broadcasts.
It can be raised to 0%. On the other hand, since both the table and the memory are used in a time-division multiplexed manner, there is only one table and its memory capacity only increases in proportion to N (since the bitmap width is N). Also, the number of output buffers only increases in proportion to N. Since there is only one table, managing the table is easy. Therefore, according to the present invention, it is possible to realize a broadcast packet switch that is efficient and has a small hardware scale, and that is simple to control and easy to manage.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a broadcast packet switch according to an embodiment of the invention. FIG. 2 is a block diagram of a conventional broadcast packet switch. FIG. 3 is a block diagram of a conventional broadcast packet switch. FIG. 4 is a diagram showing an example of a packet format. In the figure, 104... time division multiplexing unit, 105... bus, 107
...Table, 111.112.113...Gate, 121, 122.123...Memory, 140...
Each CPU is shown.

Claims (1)

[Claims] (1) A broadcast packet switch that has multiple incoming lines and multiple outgoing lines and outputs a packet from one incoming line to any one of the outgoing lines based on the identification information of the packet, and the packet switch means for time-division multiplexing packet signals and identification information of the packets from a plurality of incoming lines; means for broadcasting the time-division multiplexed packet signals to all outgoing lines; and means for broadcasting the broadcasted packet signals. a first memory means for storing and outputting to the plurality of outgoing lines for each outgoing line, and transmitting the packet to each of the plurality of outgoing lines in accordance with the identification information of each packet. a second memory means for holding information on whether to output or not; and from the time-division multiplexed identification information, each of the plurality of outgoing lines held in the second memory has the identification information. It is characterized by comprising means for reading the information indicating whether or not the packet should be output, and means for controlling whether or not the packet is stored in the first memory means according to the read information. Broadcast packet switch.