KR100238450B1 - Switch having many outlet for transferring multi-data - Google Patents

Abstract

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a switch and a packet routing method having multiple outlets for large data transmission. In the related art, there are many natural losses occurring on a communication line during packet loss occurring in a process of transmitting data at high speed. Therefore, in the present invention, when there are many packets transmitted to the same destination in interconnection of communication lines, in order to reduce packet loss due to packet collision, not only one destination but several outlets are obtained. In order to send a large amount of data to the same destination at the same time, by using an algorithm that extends the SE connection constituting the switch and the routing tag assigned to each packet in the extended switch, When the amount of data to be transmitted is applied to video communication, etc., it provides a superior communication service.

Description

Switch with Multiple Outlets for Mass Data Transmission and Packet Routing Method Using the Switch

The present invention relates to a switch and a packet routing method having multiple outlets for large data transmission. In detail, the present invention relates to a large-capacity data which allows packets passing through a network to be quickly and accurately delivered to a desired destination in accordance with a large-scale network processing requirement. A switch and a routing method of a packet having multiple outlets for transmission.

Until now, researches and inventions on how to minimize data loss and packet propagation delay time have been continuously made in high speed packet transmission networks. In the high speed data transmission, the method of satisfying both the conditions of no data loss and no time delay is very difficult with current technology, and it is a difficult problem to solve in the future.

The communication network consists of several switches, and the communication network itself can be viewed as a set of switches. And processing at the switch became a measure of overall data transfer efficiency. That is, the overall communication efficiency is determined by the performance of the switch that works with the communication medium. Types of switches can be divided into blocking switches and non-blocking. The blocking switch does not have a path assignment from the input end of the switch to the output end, and this is called blocking. On the contrary, the non-blocking switch is a path assignment from any input terminal to any output terminal. Say the switch.

The advantages and disadvantages of these two switches are: Although blocking switches have limitations in routing algorithms, routing is done in a very simple way inside the switch, which incurs the inevitable packet loss due to routing problems. However, in the case of non-blocking switch, the routing algorithm used for routing the packet is more complicated than the blocking switch, but it has the advantage of reducing packet loss during routing. The non-blocking switch has an unlimited number of input and output terminals connected to one switch element (SE), whereas the SE of the blocking switch has two input terminals and two output terminals.

Conventionally, four types of interconnection networks having multiple outlets have been proposed using blocking in the form of banyan.

The simplest interconnect is the Multiple Banyan Switch Fabric (MBSF), which connects several Bayesian interconnections in parallel to achieve multiple outlets. This method handles the packet by allocating the incoming packet to each of the connected Bainian interconnection networks in parallel. The improvement of the packet transmission rate is based on the structure of the interconnect network itself. The switch is more affected by the load balancing algorithm that assigns to each Bayesian interconnection network. However, the switch's delay time, which represents the time a packet is sent from the input to the output of the switch, is the same as a Bayan switch with a single outlet.

The second proposed switch with multiple outlets is an expanded Banyan Switch Fabric (ESBF) switch that is configured similarly to a delta switch by increasing the number of switches present in one Bayian switch. Because this switch uses one Bayian switch, the switch has the same delay time as the existing Bayesian switch even though the switch is configured to have multiple outlets. In other words, the number of stages of an EBSF switch is the same as that of a conventional Bayan switch, and the number of switches used in one stage is multiplied by the number of multiple outlets that we want to obtain. It becomes the number. However, the number of switches used in all stages, such as delta switches, is not the same, so they have fewer switches than delta switches. In the Bayesian switch, the number of switches per stage was n / 2 when the number of input ports was n. In EBSF, the maximum number of switches in one stage, that is, the number of switches in the output stage, is the number of multiple outlets to be obtained. Let k be k * n / 2.

The third proposed switch with multiple outlets is a Tandem Banyan Switch Fabric (TBSF) switch in which several Bayian switches are connected in series. The TBSF switches thus made are known to be best suited for current switch systems of Asynchronous Transfer mode (ATM). This switch is composed of a number of Bayesian switches that are as many as the number of outlets to be obtained. However, if two packets want the same output port on one switch, collisions are inevitable. Randomly select only one of the two packets that caused the collision and route to the correct output port, and add one bit to the routing tag bit to the wrong port. Mark it as a "dead packet" to indicate that the packet is being sent. The reason for marking the packet as "dead packet" is to re-route from the new serially connected switch to the correct output port. It is also the reason why the switch wants to reduce the traffic of the currently passing switch by not routing the packet marked "dead packet" before reaching the new switch. For this reason, there is a large time difference between the multiple packets having the same destination in the TBSF switch and the time to reach the destination, which is a disadvantage of the TBSF switch.

The fourth proposed switch is a switch called PBSF (Piled Banyan Switch Fabric), which has a similar transmission rate as that of TBSF, and a packet delay time of the switch structure is smaller than that of TBSF. This switch consists of a switch element whose switch element has a 2x4 port rather than a 2x2 crossbar switch structure like other switches. This means that the switch is composed of four input ports while two input ports, and the overall structure is a three-dimensional structure like MBSF. In other words, the PBSF is a stacked structure of several Bayesian switches. Although PBSF is configured in three dimensions like MBSF, there is a big difference in that the performance of the switch is not affected by the method of allocating traffic to the switch, that is, load balancing, as in MBSF. Two of the four ports, the number of output ports on each switch in the PBSF, are the ports that connect to the next stage of a normal Bayesian switch, while the other two ports are difficult to route normally due to two packet collisions occurring on that switch. One randomly selected packet is routed to the correct output port to the next stage on the same layer, and the other packet is used as a port to another Bayian switch on the lower layer. In this way, the PBSF has a similar rate as TBSF, and the average arrival time of packets arriving at the correct output port is shorter than that of TBSF since no retransmission is performed for the stages that are correctly transmitted.

As described above, the natural loss occurring on the communication line among the packet loss occurring in the process of transmitting data at a high speed can be solved by continuously supplementing the communication line and improving the communication protocol using the same. There is a problem in that a separate method must be taken to prevent the loss on the switch that is responsible for connecting to other communication lines.

The present invention provides a plurality of outlets instead of having only one destination to reduce packet loss due to packet collision when there are many packets transmitted to the same destination in interconnecting communication lines to solve the conventional problem. In order to achieve such a technology, a switch system for a large-capacity data transmission is provided by providing a switch using an algorithm for extending an SE connection constituting a switch and an algorithm for assigning a routing tag to each packet in the extended switch. Its purpose is to improve performance.

1 is a structure diagram of an extended clos switch fabric (ECSF) switch for large data transmission according to the present invention;

2 is a routing algorithm in an ECSF switch according to the present invention.

3 is a structure diagram of a Tandem Clos Switch Fabric (TCSF) switch for mass data transmission according to the present invention;

4 is a routing algorithm in a TCSF switch according to the present invention;

* Description of the symbols for the main parts of the drawings *

1,2,3: Non-blocking switch 4,5,6,7: Centralizer

10: input stage 20: intermediate stage

30: output stage

Two switches according to the present invention for achieving the above object uses a non-blocking switch. Therefore, the routing tag generated in the developed switch follows the destination-oriented routing method. The destination-oriented route assignment method can be obtained by a simpler method than the route assignment tag obtained by calculating the addresses of the input and output terminals of several route assignment switches. In order to handle the collision of the packet that occurred in the routing process of the packet on the switch with the acquired routing tag, only one packet among the collided packets is transmitted to the desired destination, and the remaining packets are transmitted through the bypass path. It uses TCSF switch technology and uses the ECSF switch's packet transmission method, in which all packets coming into the input stage are composed of multiple switches without collisions, so that the packets are transmitted using a single independent switch.

The present invention solves the problem of the conventional switch that accommodates only one packet at a time during data transmission by increasing the amount of data that can be accommodated at a time in a medium for transmitting a large amount of data.

In the present invention, a non-blocking switch having a large number of input terminals and output terminals is used, and a Clos switch, which is widely known as a non-blocking switch, is used, but the number of terminals to which the Clos switch can be connected is very large. The large number of terminals can be used to import a lot of data into the switch at once, and all the data entered into the switch can be routed to the desired destination. However, when there is more than one packet that wants the same destination, collision between the packets cannot be avoided, resulting in packet loss. One way to avoid this is to have multiple outlets with the same destination address, which can accommodate more packets than a switch with one outlet. This allows the use of communication media that can transfer large amounts of data quickly, while avoiding packet loss due to lack of switch capacity. Having multiple outlets does not require a tag for multiple routing. Packets with the same destination all have the same routing tag. In TCSF, when a collision occurs, a large number of packets are sent to the desired destination by sending one packet to the desired outlet and the other packets to the next outlet via the bypass. ECSF uses switches that extend the original switch structure by the desired number of outlets. When the incoming packet is transmitted to the next switch, it is spread over the input of the next expanded terminal, and finally, the packets arriving at the final output terminal are transmitted to the output terminal having the same outlet.

Hereinafter, with reference to the accompanying drawings, the present invention will be described in detail.

인 k×n개의 스위치를 구비하는 입력 스테이지(10)와; 각 스위치의 입력 포트와 출력 포트의 수가 n×n 포트로 이루어지는 k×m개의 스위치를 구비하는 중간 스테이지와(20), 각 스위치의 입력포트와 출력포트의 수가 m×m포트로 이루어져 총 출력 포트의 수가 k×n×m개인 출력 스테이지(30)로 구성된다.1 is a structure diagram of an extended closed switch fabric (ECSF) switch for large data transmission according to an embodiment of the present invention. As shown in FIG. Switches with multiple outlets, each switch having an input port and an output port

An input stage 10 having k × n switches; Intermediate stage with k × m switches 20 of which the number of input ports and output ports of each switch is composed of n × n ports, and the total output ports of which the number of input and output ports of each switch consists of m × m ports. The output stage 30 is a number of k x n x m.

ECSF (Extended Clos Switch Fabric) according to the present invention is a switch that extends by a multiple of the number of overlapping outlets to obtain a switch in each stage in the existing Clos switch. The ECSF consists of three fixed stages, like the existing Clos switch, with a redundant path from input to output as shown. The ECSF has three parameters and is represented by the following symbol, ECFS (n, m, k). The parameters n and m of the switch used have the same meaning as the parameters of the Clos switch, where n × m, the product of two integers, is the number of input ports given to the ECSF switch, and the last parameter k is used. K x n x m is the number of ECSF output ports.

이다. 따라서 제안한 스위치에 입력되는 모든 입력 포트의 수는 입력 스테이지에 존재하는 스위치의 갯수에 각 스위치마다 할당된 입력 포트의 수를 곱한

이 된다. 이것은 기존의 Clos 스위치에 존재하는 입력 포트의 수보다는 많거나 같은 수의 입력 포트가 된다. 다음 스테이지인 ECSF의 중간 스테이지(20)에서는 k×m개의 스위치와 각 스위치는 기존의 Clos 스위치과 같은 n×n 포트로 구성된다. 그리고 마지막 스테이지인 출력 스테이지(30)에서는 상기 입력 스테이지(10)와 같은 수의 스위치을 갖고 있지만, 각 스위치에 연결된 입력 출력 포트의 수는 기존의 Clos 스위치과 같은 m×m포트로 구성되어 있어, 제안한 스위치의 총 출력 포트의 수는 k×n×m개가 된다. 그러므로 이와 같이 구성한 ECSF 스위치는 Clos 스위치보다 k배 많은 배출구를 갖게 된다. 각 스테이지 사이의 연결 규칙에는 전체적으로 Clos 스위치과 같은 셔플(shuffle) 연결을 사용한다. 스테이지간의 연결 규칙은 각 스테이지에 존재하는 k개의 스위치를 하나의 덱(deck)으로 하면, 입력 스테이지에는 n개의 덱(deck)이 존재하고 중간 스테이지에서는 m개의 덱(deck)을 형성할 수가 있다. 이렇게 형성된 덱(deck)을 셔플(shuffle)함으로써 스테이지간을 연결한다.First, the input stage 10 is k × n as shown, and the number of input ports and output ports connected to each switch is

to be. Therefore, the number of all input ports input to the proposed switch is multiplied by the number of input ports assigned to each switch by the number of switches present in the input stage.

Becomes This is the same or more input ports than the existing Clos switch. In the intermediate stage 20 of ECSF, which is the next stage, k × m switches and each switch are composed of the same n × n port as the existing Clos switch. In the final stage, the output stage 30 has the same number of switches as the input stage 10, but the number of input output ports connected to each switch is composed of the same m × m ports as the existing Clos switch. The total number of output ports is k x n x m. Therefore, the ECSF switch configured in this way has k times more outlets than the Clos switch. The connection rule between stages uses shuffle connections such as the Clos switch as a whole. The connection rule between stages is that if one switch has k decks in each stage, there are n decks in the input stage and m decks in the intermediate stage. The decks thus formed are shuffled to connect the stages.

The ECSF switch maintains the original stage of the Clos switch, so there is no delay between packets reaching the same destination as in the TCSF switch. Since the number of stages of the switch is the same as that of the original Clos nonblocking switch, the path assignment method of the existing nonblocking switch can be used as it is in forming the route assignment tag. However, configuring a switch to have multiple outlets is much more difficult than a TCSF switch.

2 illustrates a packet routing algorithm on the ECSF switch.

Routing in ECSF follows a self-routing route assignment around a destination address similar to the Clos switch. Therefore, the address used in routing only specifies the output ports of the intermediate stage and the output stage. However, since the switches in the input stage have fewer input ports and more output ports, a switch structure different from the conventional Clos switch can be used to improve the transfer rate at the input stage. Therefore, the switch used in the input stage of the ECSF switch is used in the output portion of the knockout switch to use a switch that performs only a shift function. The movement from the input port to the output port at the switch of the input stage uses only mod operation to route the packet to the input port of the intermediate stage without collision of the packet at the input stage. Since the number of output ports of the input stage is greater than or equal to the number of input ports, the probability given to the input port of the intermediate stage is greater than or equal to the probability given to the input port of the input stage.

3 is a switch system for high capacity processing according to another embodiment of the present invention. The second proposed switch with multiple outlets is called TCSF (Tandem Clos Switch Fabric). As shown in the figure, the non-blocking switch is connected in series to have multiple outlets, and each switch includes a plurality of non-blocking switches (1, 2, 3) having a plurality of input ports and output ports connected in series. ; A concentrator (4, 5, 6, 7) for receiving a packet that reaches the desired destination each time one switch ends to send the packet that passed the first switch and the packet that passed the last switch to the desired destination. It is provided.

The TCSF is a switch having multiple outlets by connecting a Clos switch, which is a non-blocking switch like the ECSF, in series. The switch represented by TCSF (n, m, k) means the number of input ports that can be processed by the TCSF by multiplying the two previous integers. The parameter k used at the end is the number of redundant outlets to be obtained through the proposed switch.

The configuration of TCSF is to connect the existing Clos switch in series. In this case, the second switch (2) to the k (3) which is overlapped is not a three stage switch like the existing Clos switch, but an intermediate stage and an output. It is a form that connects switches composed of stage only in series.

Since the input stage of the conventional Clos switch is a stage for using a value by a random function, the duplicated switch of the TCSF switch does not affect the transmission of packets, and thus can be eliminated. And the routing marker that had the information of this removed stage can be used as a bit indicating the information of the "dead packet" for the packet arriving from the second switch. In this way, TBSF with the Bayesian switch, a blocking switch, added a new "bit" to indicate "dead packet", but TCSF used a bit more for the routing marker. Without the addition of), it has the advantage that it can be used in the proposed TCSF switch only by the routing assignment mark used in the existing Clos switch.

The connection between stages uses a shuffle connection in which one switch in the stage forms a deck in the form of a shuffle, as in the conventional Clos switch. In addition, the routing method for the packets coming into the switch uses a self-routing method that routes the destination address as the ECSF method. The routing at the input stage of the first switch is routed using the same randomized algorithm as in the existing Clos switch. However, when routing a packet, collisions are inevitable if two or more packets want to route to a common output port.

4 is a routing algorithm for transmitting a packet on a proposed TCSF switch.

If a collision occurs, select only one packet among the collision packets and route to the correct output port, and mark the first bit of the routing tag bits as "dead packets". And rerouting only packets marked as "dead packets" in a switch consisting of only an intermediate stage and an output stage without an input stage, resetting the tag bits marked as "dead packets" and again returning the packet to the desired destination address. Attempts to route the packet correctly in a switch with multiple outlets for large data transfer.

More specifically, in the event of a collision, only one of the packets in the collision is selected and routed to the correct output port, and the remaining packets are "dead packets" with three routing tag bits ( The first bit of the bit is displayed. Only those packets marked "dead packets" are rerouted in a switch consisting of only an intermediate stage and an output stage without an input stage. At this time, reset the tag bit marked as "dead packet" and try to route the packet to the desired destination address again. If the packets collide again, mark the first bit as "dead packet" as described above, and then try the correct routing again on the next connected switch. This method can be repeated k times as set in the parameter. Therefore, we can perform correct routing in the proposed switch if there are less than k packet collisions in the switch with k overlapping outlets.

Implementing a switch's topology in a TCSF switch is very simple, but the technique of handling collided packets on the switch requires a very complex technique. The switch configuration connects multiple nonblocking switches in series and adds a concentrator that accepts packets arriving at the desired destination at the end of one switch. It transmits the packet passing through to the desired destination. At this time, packet delay occurs due to the characteristics of the switch configuration and the steps of the algorithm that must pass through the routing between the arriving packets.

According to the present invention, by using a technique for modifying a routing tag to select a bypass path, even if there is a defect, it can be allowed, and in the ECSF switch configuration, it is possible to obtain a result of overlapping several switches. It is possible to obtain a switch phase capable of maintaining the number of stages of the switch. The effect of the invention obtained by doing this can improve the speed of packet transmission.

Claims (3)

In the non-blocking switch, the switch consists of three stages: an input stage, an intermediate stage, and a final stage. The number of input and output ports on each switch

An input stage including k × n switches; An intermediate stage including k × m switches each having an input port and an output port number of n × n ports; A switch having multiple outlets for mass data transmission, characterized in that the number of input ports and output ports of each switch consists of m × m ports, and an output stage having a total number of output ports k × n × m. n, m is any integer) In a switch having multiple outlets by connecting a non-blocking switch in series, A plurality of non-blocking switches having a plurality of input ports and output ports in each switch and connected in series in multiple stages; And a concentrator for receiving packets passing through the first switch and packets passing through the last switch to a desired destination each time a switch is finished to receive a packet reaching the desired destination. Switch with multiple outlets. A method of routing packets in a switch having multiple outlets by connecting non-blocking switches in series, If a collision occurs, select only one packet among the collision packets and route to the correct output port, and mark the first bit of the routing tag bits as "dead packets". Wow; Rerouting only packets marked " dead packets " in a switch comprised of only an intermediate stage and an output stage without an input stage; Resetting the tag bit marked as "dead packet" and attempting to reroute the packet to a desired destination address again. The method of rerouting a packet in a switch having multiple outlets for a large data transfer. .

Images