CN1292567C - IP Ring Distributed Bandwidth Processing Method - Google Patents

Abstract

The present invention relates to an IP loop distribution type bandwidth processing method, which relates to the technology of electric communication. In the IP loop distribution type bandwidth processing method, each node in an IP loop which uses a double-loop type topological structure is definitely accessed and controlled; a flow control packet controls the transmitting flow of data according to the information of a congestion node on the IP loop and the relative position thereof. Flow control and mutual transmission of flow information are carried out on each node by nearly-free head of line block and a fairness algorithm with weight. The problem of head of line block is solved. The present invention has the advantages that broadband resources are rationally used, and the bandwidth utilization is improved; the fairness algorithm with weight is used, and therefore, bandwidth allocation is reasonable and has strong practicability. The present invention is suitable for the lossless transmission of data.

Description

IP Ring Distributed Bandwidth Processing Method

technical field

The invention relates to electric communication technology, in particular to an IP ring distributed bandwidth processing method.

Background technique

CISCO proposed DPT technology, namely Dynamic Packet Transport, which is the fourth step in CISCO's optical interconnection strategy. Its core is target node stripping and bandwidth multiplexing, but DPT technology has the feature of head of line blocking. weakness:

DPT's SRP-fa algorithm restricts the "fairness" principle in detail, but does not fully implement bandwidth reuse. When the SRP-fa fairness algorithm restricts the uplink traffic of nodes, it only limits the total traffic. Destination nodes are not considered. As shown in Figure 1, assume that node 1, node 2, and node 3 all send data packets from the outer ring to node 4. If node 3 is congested, it will send a flow control packet with flow limit information to node 2 through the inner ring. That is, the announcement packet, node 2 then copies the flow control packet and sends it from the inner ring to node 1, following the principle of the shortest path, the control packet is sent in the opposite ring direction of the corresponding data packet path, the fairness algorithm SRP executed on node 1 and node 2 -fa limits the total uplink traffic of this node according to the traffic limit information received, the purpose is to relieve or alleviate the congestion of node 3, so that all uplink traffic of node 1 in the outer ring will be restricted; assuming that node 1 sends to The data packets of nodes 2, 3, and 4, among these data packets, only the data packets sent to node 4 are responsible for the congestion of node 3, and the data packets sent to nodes 2 and 3 have no "contribution" to the congestion of node 3 ", in SRP-fa, due to the congestion of node 3, the traffic sent from node 1 to nodes 2 and 3 is restricted, resulting in head-of-line blocking. This processing method is unreasonable and wastes bandwidth.

Contents of the invention

The purpose of the present invention is to provide an IP ring distributed bandwidth processing method for reasonably utilizing broadband resources.

This IP ring distributed bandwidth processing method is characterized in that: it comprises the steps:

A. Node stripping: In the IP ring, for unicast packets, the destination node is stripped; for multicast packets, the source node is stripped;

B. Support multiple priorities:

The packets of the IP ring are mainly divided into control packets and data packets. The format of the packet header is the same, and the packet header contains the following fields in turn: TTL, RI, MODE, PRI, and P;

TTL: Similar to the TTL field of IP packets, L3 will set the initial TTL field value according to the address of the destination node when the packet is connected to the ring, indicating the logical distance between the source node and the destination node;

RI: Indicates whether the current packet is transmitted on the outer ring or the inner ring;

MODE: Indicates the type of the current packet; currently available types include: flow control packet, topology packet, IPS packet and data packet, etc.;

PRI: Indicates the priority of the packet of the IP ring;

P: Baotou check digit, odd check for TTL, RI, MODE, PRI;

The packet format of the IP ring includes: header, DESTINATION NODE ADDRESS, SOURCE NODE ADDRESS, PAYLOAD, FCS;

Among them, DESTINATION NODE ADDRESS: destination node address, all nodes on the ring are configured with a unique logical address;

SOURCE NODE ADDRESS: source node address;

PAYLOAD: Payload, its length is variable, in bytes, usually a MAC frame;

FCS: frame check sequence; perform CRC32 check calculation on the part of the IP ring packet except the header and FCS itself;

The format of the flow control packet of the IP ring is as follows: Baotou, SOURCE NODE ADDRESS, TTL_TO_CONGEST, BAND-ALLOW, FCS;

SOURCE NODE ADDRESS: the address of the node sending the bandwidth control information;

TTL_TO_CONGEST: the logical distance from the nearest congested node to this node;

BAND-ALLOW: bandwidth control information;

The PRI field of the IP ring header indicates the sending priority of the packet. When the payload is a MAC frame, the packet priority is mapped from the VLAN priority of the MAC frame and the TOS priority of the IP packet in the MAC frame according to certain rules. 8 priorities; different priorities have different scheduling policies, and packets with higher priorities on the IP ring are transmitted first;

The TTL_TO_CONGEST in the flow control packet indicates the logical distance from the nearest congested node to the current node. When the node detects its own congestion, it sets the TTL_TO_CONGEST field of the flow control packet sent upstream to 1; if the non-congested node receives For an announcement packet, add 1 to the value of the TTL_TO_CONGEST field in the received announcement packet and assign it to the TTL_TO_CONGEST field that sent the announcement packet. If the congested node receives this announcement packet, it will reset the TTL_TO_CONGEST of the announcement packet to 1;

C. Congestion and flow processing: In the IP ring system, a fair algorithm is used to calculate the available bandwidth, so that the entire ring network forms a closed negative feedback loop with flow mutual control and distributed calculation; the flow control packet is based on the information of the congested nodes on the IP ring and Its relative position controls the flow of data transmission;

The TTL domain is subtracted by 1 at each intermediate forwarding node, and the unicast packet will be stripped by the destination node. When some data packets sent by the upstream node are stripped by the node, the node will update A large amount of L3 data is sent to the ring; the multicast packet is stripped by the source node. If both the source node and the destination node fail after the IP ring packet is on the ring and before reaching the destination node, it will be discarded when the TTL is reduced to 0. Forced to be stripped from the ring; in the congestion and traffic processing method, the SOURCE NODE ADDRESS field of the announcement packet indicates the address of the most congested node on the IP ring, and the corresponding BAND-ALLOW indicates the send_rate of the most congested node on the IP ring network;

The fairness algorithm adopts the weight fairness algorithm for each node on the ring network. According to the importance of each node and the required bandwidth flow, the corresponding weight coefficient is configured when it is initially enabled. In the calculation, the send_rate and BAND-ALLOW of each node are send_rate and BAND-ALLOW after the normalization calculation of the corresponding weight coefficient;

The greater the weight coefficient, it indicates that the importance of the node is higher. When the ring network is congested, it should enjoy more IP ring bandwidth than nodes with small weight coefficients, and the bandwidth they enjoy is proportional to the node weight value;

The maximum number of nodes supported by the IP ring is 64;

The maximum value of the TTL domain is twice the number of nodes supported by the IP ring network;

The check polynomial of the check calculation performed by the frame check sequence is:

X ³² +X ²⁶ +X ²³ +X ²² +X ¹⁶ +X ¹² +X ¹¹ +X ¹⁰ +X ⁸ +X ⁷ +X ⁵ +X ⁴ +X ² +X+1:

The above-mentioned IPR implements high and low priorities, and there is a configurable parameter as the threshold for dividing high and low priorities; L3 classifies the data packets of the upper ring into the high-priority Transmit Buffer and the low-priority Transmit Buffer according to the priority. In the priority Transmit Buffer, MAC places the upstream data packets that need to be forwarded in the high-priority Transit Buffer and low-priority Transit Buffer according to the priority; there are 4 types of data packets that need to be sent from the Tx interface of the node: high and low priority The data packets in the Transit Buffer and the data packets in the high and low priority Transmit Buffer; in the IP ring, the packet sending scheduling strategy follows the following principles:

High-priority packets should be sent prior to low-priority packets; for services that have been connected to the ring, the data packets in the Transit Buffer should not be discarded; the general priorities of the four types of data packet scheduling in IP ring nodes are:

High-priority Transit packets, forwarding high-priority packets from upstream nodes on the ring;

High-priority Transmit packets, high-priority packets on the ring from this node;

Low-priority Transmit packets, low-priority packets from the local node on the ring;

Low-priority Transit packets, forwarding low-priority packets from upstream nodes on the ring;

The reserved bandwidth is configured for high-priority data packets. The data packets in the high-priority Transit Buffer are always sent unconditionally first, but the data packets in the high-priority Transit Buffer are sent conditionally first: if the cache of the low-priority Transit Buffer When the dangerous high threshold HI_THRESHOLD is reached, the node must stop sending high-priority Transit packets, immediately send low-priority Transit packets, and the MAC outputs a STOP_HIGH signal to L3, instructing to stop sending high-priority Transmit packets;

Before the low-priority Transit Buffer depth reaches the dangerous low threshold LO_THRESHOLD, the low-priority Transit packet is sent before the low-priority Transit packet, but once the dangerous low threshold is reached, the sending of the low-priority Transit packet must be stopped immediately and the low-priority Transit packet will be forwarded immediately. Priority Transit packets; low-priority Transmit packets will also stop sending after the flow exceeds the limit of the bandwidth processing method, and the MAC outputs a STOP_LOW signal to L3, indicating to stop sending low-priority Transit packets;

The depth of the high-priority Transit Buffer is 2-3 maximum packet length MTU, the maximum packet length MTU is Max Transport Unit, the maximum packet length MTU in the IP ring is 9216 bytes; the low-priority Transit Buffer must be large enough ;The dangerous low threshold of the low-priority Transit Buffer is about half of the entire Buffer size, and the dangerous high threshold is set to the entire Buffer size minus 3 maximum packet length MTU sizes.

The principle and beneficial effects of the present invention are: IP ring is a kind of IP packet ring network transmission technology, which is used for the optimized transmission of IP packets in the metropolitan area network, and the IP ring network system is a new type of transmission system, which combines the router The characteristics of high bandwidth utilization efficiency and the advantages of large bandwidth and strong self-healing ability of optical fiber network, IP ring uses the target node stripping technology to fully optimize the local bandwidth and realize bandwidth multiplexing, that is, the comprehensive bandwidth utilization rate of the ring network is higher than that of In the case of non-multiplexing, at the same time, according to certain rules, each node on the ring network can obtain a reasonable and fair amount of access bandwidth. The IP ring uses the same dual-ring topology structure as all ring technologies such as Token Ring and FDDI. As shown in Figure 1, a balanced bidirectional ring topology structure is used. The two rings are called the inner ring and the outer ring respectively. The two rings transmit data packets and control packets concurrently, and the related data packets and control packets are transmitted on the opposite ring direction. For the data packets of the outer ring, the related control packets are sent in the opposite direction of the inner ring. The IPR topology packet is a control packet that completes the ring network topology map. A control packet to complete protection switching, IP ring bandwidth control packet, a control packet to complete the bandwidth limit information transmission function; in normal (non-switching) working mode, there are two data paths between any two nodes, the data packet The route can choose the outer ring route, as shown in Figure 1, the data packet sent by node 1 to node 3 can choose the outer ring route node 1->node 2->node 3, or choose the inner ring route, as shown in Figure 1 Among them, the data packet sent by node 1 to node 3 can choose the inner ring route node 1->node 6->node 5->node 4->node 3, which ring to choose is determined by the L3 layer, generally following the shortest path principle, According to this principle, the data packet sent by node 1 to node 3 will choose the routing node 1->node 2->node 3; as a shared medium, it is necessary to perform certain access control on each node on the ring to ensure the Each node on the Internet shares the ring bandwidth reasonably and fairly, but unlike Token Ring and FDDI, IP Ring does not use tokens for centralized access control. It completes flow control by executing a distributed bandwidth processing method on each node. Special flow control packets are used to transmit flow information between nodes, and the distributed bandwidth processing method on nodes is directly implemented on hardware, such as ASIC and FPGA; the IP ring is actually a flow control feedback loop. Since all bandwidth calculations, traffic mutual control, and congestion processing are directly completed by the hardware of each node without software control, the response speed is greatly improved and the convergence time from traffic mutation to traffic fairness is shortened.

In the present invention, the TTL_TO_CONGEST in the flow control packet indicates the logical distance between this node and the nearest congested node, and once the node detects that it is congested, it will set the TTL_TO_CONGEST field of the flow control packet sent upstream to 1, no If a congested node receives an announcement packet, it will add 1 to the value of the TTL_TO_CONGEST field in the received announcement packet, and assign it to the TTL_TO_CONGEST field that sent the announcement packet. The example is as follows, as shown in Figure 1. Assuming that node 4 is congested, node 4 1, 2, 3, 5, and 6 are not congested, so in the announcement packet sent by node 4 to node 3, SOURCE NODE ADDRESS is equal to 4, TTL_TO_CONGEST is equal to 1, BAND-ALLOW is the send_rate of node 4, and node 3 receives this After the announcement, copy an announcement with the same content, then add 1 to TTL_TO_CONGEST to change it to 2, send it to node 2, and node 2 will do the same processing... After receiving the announcement, node 1 uses the information in the announcement packet To limit your own uplink traffic, first ensure that the sum of the low-priority bandwidth sent to nodes beyond 4 but not including node 4 is not greater than the received BAND-ALLOW, that is, the send_rate of node 4, and then check the TTL_TO_CONGEST domain of the received announcement , and found to be 3, we know that the logical distance to the nearest congested node is 3, so node 1 sends Transmit packets to node 2, node 3, and node 4 without restriction, and L3 of node 1 will send to The TTLs of the data packets of these 3 nodes are set to 1, 2, 3 respectively. In this way, the congestion of node 4 limits the traffic sent by node 1 to nodes far away from node 4, but does not affect the bandwidth sent to nodes 2, 3, and 4, which solves the problem of head-of-line blocking and improves bandwidth utilization;

In the present invention, a weight fairness algorithm is added. When the ring network traffic is fair, the bandwidth utilization quantity of each node is proportional to the configured node weight, so that the bandwidth allocation is more reasonable, and the bandwidth flow of each node on the ring network is in line with the actual use. Need, improved the practicability of the present invention; The number of nodes supported by IP ring network is at most 64, the maximum value of TTL domain adopts twice of the number of nodes supported by IP ring network i.e. 128, although the priority domain PRI represents 8 Priority, in the IP ring of the present invention, only adopt high and low two priority, differentiate with priority threshold, and the depth in the high priority Transit Buffer is 2-3 maximum packet length MTU, low priority The dangerous low threshold of TransitBuffer is about half of the whole Buffer size, and the dangerous high threshold is set to the whole Buffer size minus 3 maximum packet length MTU sizes, etc., which makes the present invention more reasonable and practical in practical engineering applications.

In a word, the present invention rationally utilizes broadband resources, solves head-of-line blocking problem, improves bandwidth utilization rate, uses weight fairness algorithm, makes bandwidth allocation more reasonable, has strong practicability, and is suitable for lossless data transmission.

Description of drawings

Figure 1 is a schematic diagram of a double-ring topology;

Fig. 2 is a schematic diagram of an IP ring node realization model;

Fig. 3 is a schematic diagram of IP ring connection;

Fig. 4 is a schematic diagram of a ring network application example.

Detailed ways

The present invention will be described in further detail below according to accompanying drawing and embodiment: according to Fig. 1, Fig. 2, Fig. 3 and Fig. 4, in the IP ring that adopts double-ring topological structure, each node on the ring is carried out access control, by A certain processing method completes the flow control on each node, and the nodes transmit flow control information to each other. The IP ring node needs to process the inner ring and outer ring traffic at the same time. Therefore, an IP ring node is composed of two back-to-back MAC processing chips. Process the packets of inner ring and outer ring, as shown in Figure 2, for an IPR node, there are two kinds of input traffic: the traffic received from the upstream node (dividing the upstream and downstream according to the routing path of the data packet) and the traffic on this node The traffic of the ring also has two kinds of output traffic: the traffic sent from the upstream node and destined for this node and the traffic sent to the downstream node. The MAC node is actually a packet transponder; as shown in Figure 3, the Rx and Tx of the adjacent nodes of the IP ring are connected end to end to form an IP ring network. The present invention adopts the following method to complete the distribution of the distributed flow control information on each node. calculate:

A, node stripping: traditional data rings such as token ring and FDDI ring use source address stripping and token processing to control the access to the ring network, and the data packet will go around the entire ring network for one week before being stripped. The present invention implements The destination node of the unicast packet is stripped. When the node receives the unicast data packet from the upstream adjacent node at Rx, it performs CAM matching. If the DESTINATION NODE ADDRESS of the unicast data packet is equal to the address of the node, the The data packet is sent to the HOST, that is, the L3 device, and the data packet is stripped from the ring; if the CAM match fails, indicating that the node is not the destination node, the data packet is placed in the Transit Buffer, and the flow control module Flow Control Under the control of the Module, it continues to forward to the downstream nodes. For multicast packets, the source node stripping method is implemented. The destination node stripping method makes bandwidth multiplexing possible. The total bandwidth output by the Tx interface is certain. When some upstream nodes send After the data packets are stripped by the node for the purpose, the node sends more L3 data to the ring, thus improving the bandwidth utilization rate; in addition, the reverse double-ring structure of the IP ring is also conducive to bandwidth multiplexing, because the double-ring There are always two routing paths between any source node and destination node, the inner ring and the outer ring. Generally, a shorter routing path is selected, and the logical distance of the shorter path must not be greater than the logical perimeter of the ring (that is, the total number of ring nodes). Half, assuming that the number of data packets sent by each source node to each destination node is randomly and evenly distributed, then the average logical distance of all routes is only 1/4 of the logical perimeter of the ring, if the bandwidth of all routing segments is utilized , will greatly improve the efficiency of bandwidth multiplexing.

B. Support multiple priorities:

The packets of the IP ring are mainly divided into control packets and data packets. The format of the packet header is the same, and the packet header contains the following fields in turn: TTL, RI, MODE, PRI, and P;

TTL (8bits): Similar to the TTL field of IP packets, L3 will set the initial TTL field according to the address of the destination node when the packet is on the ring, indicating the logical distance of the selected route, and each intermediate forwarding node will subtract 1 from the TTL field , the unicast IPR packet will be stripped by the destination node, and if it is a broadcast packet, it will be stripped by the source node. If both the source node and the destination node fail before reaching the destination node after the IPR packet is on the ring, it will be forcibly stripped from the ring when the TTL is reduced to 0. This is mainly to prevent the IPR packet from being transmitted infinitely on the ring. The maximum value of TTL is twice the maximum number of nodes. The current IP ring supports up to 64 nodes. The more nodes, the longer the distance between nodes, and the longer the convergence time of ring network traffic regulation.

RI (1bit): indicates whether the current packet is transmitted on the outer ring or the inner ring;

MODE (3bit): Indicates the type of the current packet; the types that can be referred to include: flow control packet, topology packet, IPS packet and data packet, etc.;

Among them, 000-011: Reserved

                                         

                                     

                                   

                                     

PRI (3bits): Indicates the priority of the packet of the IP ring;

P(1bit): Packet header parity bit, odd parity for TTL, RI, MODE, PRI;

The packet format of the IP ring includes: header, DESTINATION NODE ADDRESS, SOURCE NODE ADDRESS, PAYLOAD, FCS;

Among them, DESTINATION NODE ADDRESS (8bits): the destination node address, all nodes on the ring are configured with a unique logical address, so that the MAC on all nodes performs fast layer-2 switching according to the logical address;

SOURCE NODE ADDRESS (8bits): source node address;

PAYLOAD: payload, its length is variable, in bytes, it can be any payload, usually a MAC frame;

FCS (32bits): frame check sequence; CRC32 check calculation is performed on the part of the IP ring packet except the header and FCS itself, and the check polynomial of the check calculation performed by the frame check sequence is:

X32+X26+X23+X22+X16+X12+X11+X10+X8+X7+X5+X4+X2+X+1

The format of the flow control packet of the IP ring is as follows: Baotou, SOURCE NODE ADDRESS, TTL_TO_CONGEST, BAND-ALLOW, FCS;

SOURCE NODE ADDRESS (8bits): the node that sends out bandwidth control information;

TTL_TO_CONGEST(8bits): The logical distance between this node and the nearest congested node to this node;

BAND-ALLOW (16bits): bandwidth control information;

The other formats of the control packets are the same as the corresponding other formats of the data packets.

The PRI domain of the IP ring packet header indicates the sending priority of the packet. The packet priority is mapped by the VLAN priority of the MAC frame and the TOS priority of the IP packet in the MAC frame according to certain rules. A total of 8 priorities are supported; There are different scheduling strategies for the priority. On the IP ring, packets with high priority are transmitted first; for the sake of implementation cost, two priority levels are implemented in the IP ring, and there is a configurable parameter as the high and low priority. Low priority threshold; as shown in Figure 2, L3 classifies the data packets on the upper ring into the high priority TransmitBuffer and low priority Transmit Buffer according to the priority, and the MAC places the upstream data packets that need to be forwarded in the In the high-priority Transit Buffer and low-priority Transit Buffer; there are 4 types of data packets that need to be sent from the Tx interface of the node: the data packets in the high-priority and low-priority Transit Buffer and the data packets in the high-priority and low-priority Transit Buffer; In an IP ring, the packet sending scheduling policy follows the following principles:

High-priority packets should be sent prior to low-priority packets; data packets in the Transit Buffer should not be discarded for services that have been connected to the ring. The situation that the services that have been connected to the ring are discarded generally occurs when the low-priority Transit Buffer has been If the upstream node has a low-priority data packet that needs to be forwarded by this node, it comes from the Rx interface. The newly arrived data packet cannot be written into the low-priority Transit Buffer and has to be discarded. Therefore, preventing congestion or overflow of low-priority Transit Buffers is an important role of scheduling strategies and fairness algorithms. Packets in high-priority Transit Buffers are always scheduled preferentially at nodes, so there is no danger of being discarded.

The general priorities of the four types of data packet scheduling in IP ring nodes are:

High-priority Transit packets, forwarding high-priority packets from upstream nodes on the ring;

High-priority Transmit packets, high-priority packets on the ring from this node;

Low-priority Transmit packets, low-priority packets from the local node on the ring;

Low-priority Transit packets, forwarding low-priority packets from upstream nodes on the ring;

By sending low-priority Transmit packets first, the depth of the low-priority TransitBuffer reaches the congestion threshold. After congestion, the node sends a flow control packet with flow limit information to the upstream node, that is, an announcement packet, so that the upstream node reduces the traffic on the ring. , to alleviate or eliminate the low priority Transit Buffer congestion of the node, therefore, the low priority Transit packet is prioritized over the low priority Transit forwarding.

High-priority data packets are configured with reserved bandwidth, and data packets in the high-priority Transit Buffer are always sent unconditionally first, but the sum of the high-priority reserved bandwidth configured by each node cannot be too large, otherwise the low-priority data packets will The forwarding delay after the ring is very large, and it is easy to cause node congestion or even overflow; the data packets in the high-priority Transit Buffer are sent conditionally and preferentially: if the cache of the low-priority Transit Buffer reaches the dangerously high threshold HI_THRESHOLD, it means that the low-priority Transit Buffer Buffer is about to overflow, the high-priority Transit packet of the node must be stopped, and the low-priority Transit packet should be sent immediately to prevent the low-priority Transit Buffer from overflowing. At this time, MAC outputs STOP_HIGH signal to L3, indicating to stop sending high-priority Transit packet ;

The strategy of sending low-priority Transit Buffer data packets prior to low-priority TransitBuffer data packets is also conditional: before the low-priority Transit Buffer depth does not reach the dangerous low threshold LO_THRESHOLD, low-priority Transit packets are sent before low-priority Transit packets, But once the dangerous low threshold is reached, the forwarding of low-priority Transit packets must be stopped immediately, and the low-priority Transit packets should be forwarded immediately to avoid low-priority Transit Buffer overflow. The low-priority Transmit packet should also stop sending after the traffic exceeds the limit of the bandwidth processing method. At this time, the MAC outputs a STOP_LOW signal to L3, indicating to stop sending the low-priority Transmit packet;

Since the packets in the high-priority Transit Buffer are always forwarded first, it does not need to be too large. The depth in the high-priority Transit Buffer is 2-3 maximum packet length MTU, the maximum packet length MTU is Max Transport Unit, in IPR The maximum packet length MTU is 9216 bytes; the low-priority Transit Buffer must be large enough to avoid overflow, the specific size depends on the simulation situation, in the case of a loop bandwidth of 622Mbps and a ring diameter of 100km, the recommended Buffer capacity 256Kbyte, when the ring bandwidth or ring diameter increases, the Buffer capacity should be increased accordingly; the dangerous low threshold of the low priority Transit Buffer is about half of the entire buffer size, and the dangerous low threshold is too low to cause frequent congestion, and thus lead to low priority The packet delay is too large, and the dangerously high threshold can only be reached when the loop congestion is extremely severe. It is set to the entire Buffer size minus 3 maximum packet length MTU sizes.

C. Congestion and traffic processing: In the IP ring system, a weighted fairness algorithm is adopted to make the entire ring network form a closed negative feedback loop with traffic mutual control and distributed calculation. Its relative position controls the flow of data transmission. The flow control information BAND-ALLOW in the flow control packet is also called "announcement", and the flow control packet is also called "announcement packet". If the node that receives the announcement is not congested, then Continue to send the received announcement to the upstream node. If the node receiving the announcement is also congested, it compares the received BAND-ALLOW with its own send_rate, and chooses the smaller one to send upstream in the announcement packet. The SOURCE NODE ADDRESS field in the flow control packet indicates which node the value of BAND-ALLOW comes from. The non-congested node and the node whose send_rate is greater than the received BAND-ALLOW will copy the contents of the SOURCE NODE ADDRESS field that received the flow control packet to this node The SOURCE NODE ADDRESS field of the flow control packet sent, the SOURCE NODE ADDRESS field of the flow control packet essentially indicates the address of the most congested node on the IP ring, and the corresponding BAND-ALLOW indicates the most congested node on the IP ring network send_rate; TTL_TO_CONGEST in the flow control packet indicates the logical distance from the nearest congested node. When the node detects its own congestion, it will set the TTL_TO_CONGEST field of the flow control packet sent upstream to 1; if the non-congested node receives an announcement packet, add 1 to the value of the TTL_TO_CONGEST field in the received announcement packet, and assign it to the TTL_TO_CONGEST field that issued the announcement. In this way, each non-congested node knows the logical distance between the nearest congested node and its own node. In this way, the non-congested node can send data packets to the nearest congested node with the method of nearly no head-of-line blocking. The present invention applies the method of near-no-head-of-line blocking due to congestion, and well solves the problem of head-of-line blocking.

In the IP ring system, each node periodically sends a flow control packet, and the BAND-ALLOW in the flow control packet indicates the low-priority bandwidth that the upstream node is allowed to send and needs to be forwarded by the node. As shown in Figure 1, the node 4 The flow control packet sent to node 3 indicates the low-priority bandwidth that node 3 can send to node 4. In the IP ring network, the flow control packets are all point-to-point, and the traffic sent by node 4 to node 3 Control packets are stripped by node 3.

If all nodes on the ring network are not congested, each node can try to send low-priority data packets. At this time, the BAND-ALLOW of the flow control packet sent by each node is 16'hFFFF, which means that it does not limit the flow of upstream nodes. On-ring low-priority traffic; once a node is congested, that is, the depth of the low-priority Transit Buffer reaches a congestion threshold configured when the system is initially enabled, the node will set the sent BAND-ALLOW as its own local low-priority The upper ring rate send_rate, after the upstream node receives this traffic control announcement packet, it will adjust its upper ring low priority Transmit traffic so that it does not exceed BAND-ALLOW. In short, the low-priority Transmit bandwidth sent by the upstream node of the congested node should not be larger than the low-priority Transmit bandwidth sent by the congested node, which reflects fairness, and the uncongested node receives the restricted traffic sent by the downstream node After the announcement, the BAND-ALLOW will be copied into the flow control packet and continue to be sent to the upstream node. In this way, the entire ring network forms a closed negative feedback loop with mutual flow control and distributed computing, and finally the traffic of each node in the ring network reaches stability. and fairness, but this is just a simple fairness. For example, as shown in Figure 4, nodes 1, 2, and 3 all send data packets to 4 on the outer ring. Assuming that the bandwidth of a single ring is 600M, then node 3 is the first to congest , Node 3 will first send the announcement to Node 2, and Node 2 will copy the announcement and send it to Node 1. Under the action of the fairness algorithm, the final convergence result of the loop is that nodes 1, 2, and 3 each send 200M traffic to node 4, which is absolute fairness without weight. In actual network applications, the traffic of each node is not consistent. For example, node 3 can provide an average traffic of 500M, and nodes 1 and 2 can provide an average traffic of 300M; and the business of node 3 may be more important than that of nodes 1 and 2. Simplicity and fairness seem unreasonable.

In the fairness algorithm, the present invention configures corresponding weight coefficients for each node on the ring network according to its importance and the required bandwidth flow when it is initially enabled. In the calculation, the send_rate and BAND-ALLOW of each node are passed through the corresponding weight coefficients Normalized calculation of send_rate and BAND-ALLOW; the larger the weight coefficient, the higher the importance of the node, and it should enjoy more IP loop bandwidth than the node with a small weight coefficient during congestion. Obviously, using a fair Algorithms solve simple fairness problems. For example, each node on the ring network is configured with a weight coefficient when it is initially activated. For example, the weight coefficients configured for nodes 1, 2, and 3 are 1, 2, and 3 respectively. The greater the weight, the higher the importance of the node. Nodes with smaller weight coefficients enjoy more loop bandwidth. As shown in Figure 4, when node 3 is congested, it does not send its own send_rate to node 2, but divides send_rate by 3, which is the weight coefficient of node 3, and then sends it to 2. It is the flow control information after normalized calculation. After receiving the announcement, the fairness algorithm executed on node 2 will limit the low-priority data packets on the upper ring to not be larger than the received BAND-ALLOW multiplied by 2, that is, node 2’s Weight coefficient, similarly, the fairness algorithm executed on node 1 restricts the low-priority packets on the ring to no more than BAND-ALLOW multiplied by 1; after the bandwidth of the ring network converges, the low-priority The level bandwidth is 100M, 200M, 300M respectively, which makes the bandwidth allocation more reasonable.

Claims (9)

1. a kind of IP ring distributed bandwidth processing method is characterized in that: it comprises the steps: A. Node stripping: In the IP ring, for unicast packets, the destination node is stripped; for multicast packets, the source node is stripped; B. Support multiple priorities: The packets of the IP ring are mainly divided into announcement packets and data packets. The format of the packet header is the same, and the packet header contains the following fields in sequence: TTL, RI, MODE, PRI, and P; TTL: Similar to the TTL field of IP packets, L3 will set the initial TTL field value according to the address of the destination node when the packet is connected to the ring, indicating the logical distance between the source node and the destination node; RI: Indicates whether the current packet is transmitted on the outer ring or the inner ring; MODE: Indicates the type of the current package; currently available types include: announcement package, topology package, IPS package and data package, etc.; PRI: Indicates the priority of the packet of the IP ring; P: Baotou check digit, odd check for TTL, RI, MODE, PRI; The packet format of the IP ring includes: header, DESTINATION NODE ADDRESS, SOURCE NODE ADDRESS, PAYLOAD, FCS; Among them, DESTINATION NODE ADDRESS: destination node address, all nodes on the ring are configured with a unique logical address; SOURCE NODE ADDRESS: source node address; PAYLOAD: Payload, its length is variable, in bytes, usually a MAC frame; FCS: frame check sequence; perform CRC32 check calculation on the part of the IP ring packet except the header and FCS itself; The format of the announcement packet of the IP ring is as follows: Baotou, SOURCE NODE ADDRESS, TTL_TO_CONGEST, BAND-ALLOW, FCS; SOURCE NODE ADDRESS: the address of the node sending the bandwidth control information; TTL_TO_CONGEST: the logical distance from the nearest congested node to this node; BAND-ALLOW: bandwidth control information; The PRI field of the IP ring header indicates the sending priority of the packet. When the payload is a MAC frame, the packet priority is mapped from the VLAN priority of the MAC frame and the TOS priority of the IP packet in the MAC frame according to certain rules. 8 priorities; different priorities have different scheduling policies, and packets with higher priorities on the IP ring are transmitted first; The TTL_TO_CONGEST in the announcement packet indicates the logical distance from the nearest congested node to the node. When a node detects its own congestion, it sets the TTL_TO_CONGEST field of the announcement packet sent upstream to 1; if a non-congested node receives an announcement packet, add 1 to the value of the TTL_TO_CONGEST field in the received announcement packet, and assign it to the TTL_TO_CONGEST field of the announcement packet. If the congested node receives this announcement packet, it will reset the TTL_TO_CONGEST of the announcement packet to 1; C. Congestion and traffic processing: In the IP ring system, a fair algorithm is used to calculate the available bandwidth, so that the entire ring network forms a closed negative feedback loop with traffic mutual control and distributed calculation; the announcement packet is based on the information of the congested nodes on the IP ring and its The relative position of the control data transmission flow. 2. IP ring distributed bandwidth processing method according to claim 1, is characterized in that: described TTL domain subtracts 1 by TTL domain at each intermediate forwarding node, and unicast packet can be stripped by purpose node, when there is After part of the data packets sent by the upstream node are stripped by the node, the node sends more L3 data to the ring; the multicast packet is stripped by the source node. If the IP ring packet is on the ring, and If both the source node and the destination node fail before reaching the destination node, they will be forcibly stripped from the ring when the TTL is reduced to 0; In the congestion and flow processing method, the SOURCE NODE ADDRESS domain of the announcement packet indicates the address of the most congested node on the IP ring, and the corresponding BAND-ALLOW indicates the send_rate of the most congested node on the IP ring network. 3. IP ring distributed bandwidth processing method according to claim 1, is characterized in that: described fairness algorithm adopts weight fairness algorithm to each node on the ring network, according to each node importance degree and required bandwidth flow at the beginning of using When configuring the corresponding weight coefficient, in the calculation, the send_rate and BAND-ALLOW of each node are the send_rate and BAND-ALLOW after the normalization calculation of the corresponding weight coefficient. 4. The IP ring distributed bandwidth processing method according to claim 3, characterized in that: the greater the weight coefficient, the higher the importance of the node is, and when the ring network is congested, the nodes with smaller weight coefficients should enjoy More IP loop bandwidth, the bandwidth they enjoy is proportional to the node weight value. 5. The IP ring distributed bandwidth processing method according to claim 1 or 2, characterized in that: the maximum number of nodes supported by the IP ring is 64. 6. The IP ring distributed bandwidth processing method according to claim 1 or 2, characterized in that: the maximum value of the TTL domain is twice the number of nodes supported by the IP ring network. 7. IP ring distributed bandwidth processing method according to claim 1, is characterized in that: the check polynomial of the check calculation that described frame check sequence is carried out is: X ³² +X ²⁶ +X ²³ +X ²² +X ¹⁶ +X ¹² +X ¹¹ +X ¹⁰ +X ⁸ +X ⁷ +X ^{5 +X 4 +} X ² +X+1 8. according to claim 1 or 2 described IP ring distributed bandwidth processing methods, it is characterized in that: realize high and low two priorities in the described IP ring, have a configurable parameter as division high, low priority Level threshold; L3 classifies the data packets on the upper ring into the high-priority Transmit Buffer and low-priority Transmit Buffer according to the priority, and MAC places the upstream data packets that need to be forwarded into the high-priority Transit Buffer and low-priority Transit Buffer according to the priority In the priority Transit Buffer; there are 4 types of data packets that need to be sent from the Tx interface of the node: the data packets in the high and low priority Transit Buffer and the data packets in the high and low priority Transmit Buffer; in the IP ring, the packet is sent The scheduling policy follows the following principles: High-priority packets should be sent prior to low-priority packets; for services that have been connected to the ring, the data packets in the Transit Buffer should not be discarded; the general priorities of the four types of data packet scheduling in IP ring nodes are: High-priority Transit packets, forwarding high-priority packets from upstream nodes on the ring; High-priority Transmit packets, high-priority packets on the ring from this node; Low-priority Transmit packets, low-priority packets from the local node on the ring; Low-priority Transit packets, forwarding low-priority packets from upstream nodes on the ring; The reserved bandwidth is configured for high-priority data packets. The data packets in the high-priority Transit Buffer are always sent unconditionally first, but the data packets in the high-priority Transit Buffer are sent conditionally first: if the cache of the low-priority Transit Buffer When the dangerous high threshold (HI_THRESHOLD) is reached, the node's high-priority Transmit packets must stop sending, and immediately send low-priority Transit packets, and the MAC outputs a STOP_HIGH signal to L3, instructing to stop sending high-priority Transmit packets; Before the depth of the low-priority Transit Buffer reaches the dangerous low threshold (LO_THRESHOLD), the low-priority Transit packet is sent before the low-priority Transit packet, but once the dangerous low threshold is reached, the sending of the low-priority Transit packet must be stopped immediately. Forward low-priority Transit packets; after the flow of low-priority Transit packets exceeds the limit of the bandwidth processing method, the transmission of low-priority Transmit packets should also be stopped, and the MAC outputs a STOP_LOW signal to L3, indicating that the low-priority Transmit packets should stop sending. 9. IP ring distributed bandwidth processing method according to claim 8, is characterized in that: the depth of described high-priority Transit Buffer is 2-3 maximum packet length (MTU), and maximum packet length (MTU) is Max TransportUnit, the maximum packet length (MTU) in the IP ring is 9216 bytes; the low-priority Transit Buffer must be large enough; the dangerous low threshold of the low-priority Transit Buffer is about half of the entire Buffer size, and the dangerous high threshold is set to the entire Buffer size minus 3 maximum packet size (MTU) size.

Images