JP2003163687A - Method and device for controlling route - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To dynamically optimize traffic distribution over a plurality of routes in a network. <P>SOLUTION: A flow classifying part 1 classifies an inputted packet flow into either one of groups based on a predetermined weighting factor in each group and flow output parts 2 and 3 output the flows to different routes at every group. A band width measuring part 4 measures a band width at each group which is used in each flow among the band widths used in the respective routes. A unit band width calculating part 5 calculates a unit usage band width at every group by subtracting the measured usage band width in each group by the weighting factor which is previously assigned to each group. A weighting factor adjusting part 6 adjusts the weighting factor of each group based on the unit usage band width of the group. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a route control method and device, and more particularly to a route control method and device for dynamically distributing loads to a plurality of routes in a packet transfer network.

[0002]

2. Description of the Related Art In recent years, the communication environment has been accelerating the access line speed, that is, the transmission speed between the service subscriber terminal and the nearest subscriber accommodation device has been dramatically improved, but the core of the carrier has been improved. Network acceleration has been delayed by a number of factors, and it is possible for a small number of users to easily exhaust the core bandwidth of the network.

Under such circumstances, in order to improve utilization efficiency of network resources, traffic
Engineering (hereinafter referred to as TE: Traffic Engine
The field of ering) is being actively researched. Conventional IP
In the network, packet transfer along the route with the lowest cost to the end point (shortest route) was the mainstream,
In TE, by making the second and third routes with low cost participate in packet transfer, it is possible to avoid the concentration of load on a specific relay line and thereby improve the utilization efficiency of network resources. I am trying.

In TE, for example, the required bandwidth of the line connecting each user's base is estimated based on the measurement on the network side and the request from the user, and the route allocated to the line is determined so that the network resources can be used most efficiently. Perform optimization calculation. But this calculation is complicated,
It takes a long time to obtain an optimal solution in a large-scale network. As a result, it often happens that it takes several days or more to change the route setting due to the increase or decrease of traffic. Such a method of optimizing network resources over a long time will be referred to as static TE below.

Along with the speeding up of access lines as described above, as the traffic of the Internet, the traffic with a high burst property to an unspecified large number is rapidly increasing. That is, unlike conventional leased lines and telephones, the other party of communication is not constant, and the used bandwidth has a large difference between when data is transferred and when it is not transferred. As described above, it is difficult to efficiently use network resources only for the conventional static TE with respect to the traffic whose route and used bandwidth fluctuate greatly in a short time, and it is more sensitive to the fluctuation in a shorter time. There is a strong demand for a method of responding to and optimizing resources. Such a method is referred to below as dynamic TE
I will call it.

[0006]

However, there is a problem that the conventional load balancing method cannot be applied to such a dynamic TE. Next, the technical problems in applying the conventional load balancing method to the dynamic TE will be described. In general, in order to perform TE, the load, that is, the degree of utilization of each route of the network is measured, the measurement result is analyzed, the calculation for optimizing the resource allocation is performed, and the traffic is transferred by the new route based on the calculated result. , And three steps.

The load on the IP network is measured by a server (monitoring load information) for each path control device to be monitored, and an SNMP (Simple Network Man
It is common to obtain information remotely using the agement Protocol). However, there are many problems in applying this method to dynamic TE. In the following, problems in collecting load information will be described by taking the case of using SNMP as an example. First, it is difficult to handle overloaded links and shapers. If there is an overloaded link (link that has used up physical speed) or a shaper (device that shapes traffic to a certain bandwidth) in the middle of the route, simply obtain the flow rate of the link as load information. There is a lack of information alone. In other words, when the maximum speed is used up, it is not possible to obtain information about how much traffic is actually exceeded and how much traffic should be distributed to other routes.

In order to solve this problem strictly, it is necessary to measure the difference between the input bandwidth and the output bandwidth of each route control device for each route. Moreover, since a route control device generally has multiple input interfaces and output interfaces, and traffic is repeatedly demultiplexed and aggregated within multiple route control devices, the cause of congestion on a certain link can be traced through all routes. It is necessary to check the amount of traffic that merges and the route control table in the route control device. However, collecting these information at high speed,
Very difficult to analyze. Therefore, even if an overloaded link or a shaper exists in the middle of a route, a simple and effective method for estimating the degree of load and the amount that needs to be distributed to other routes is needed.

Secondly, there is a new complication due to the server installation. In dynamic load balancing, it is necessary to collect load information momentarily every few seconds. To collect load information frequently in a large network,
It is more efficient that a dedicated server that collects load information collects and analyzes the information collectively, rather than each path control device individually performing measurement or analysis. But this is
It introduces new complexity into the construction of networks. For example, it is necessary to consider the setting and duplication of another line for load information, and the duplication of servers.
There is a strong demand for an efficient method of collecting information without using such a server.

Third, there is an increase in the number of management targets.
In order to properly bypass each route, it is necessary to measure the load for each route to all route control devices in the network. However, in a large-scale network, the number of routes and the number of route control devices increase dramatically, so it is difficult to measure and analyze the results of each waypoint of these routes. As a result, it takes a long time to perform the optimization calculation with respect to the fluctuation of the load, and it is difficult to apply it to the dynamic TE. There is a need for a method of effectively distributing loads with a small number of management targets.

Fourth, it is difficult to deal with a failure. Dynamic TE requires the ability to respond instantly to network failures. Therefore, when the amount of traffic on a certain route is small, it is necessary to determine in a short time whether the cause is a link failure, a route control fault, or simply because it is free. This is because it is preferable that the traffic be actively flowed to the free route if it is free, and conversely, if there is a failure, it is desirable not to use the route.

However, in an actual network, the SNM
It is difficult to isolate such obstacles using P alone. For example, if there is a repeater or a switch that does not have the SNMP management function on the route, it is not possible to detect the failure that occurred in that part. In addition, the cause of the failure is not limited to the interface failure, but various factors such as the occurrence of a loop in the transient state of the route control can be considered. Therefore, in order to accurately detect a failure on the route control, in addition to monitoring the states of all elements involved in packet transfer including such repeaters and switches, each route control device holds the state. It is necessary to acquire the route control information (information exchanged by the route control protocol) and the route control table.

Further complicating the problem is that a plurality of routing protocols are often used together even within the same network. That is, static route setting by manual operation and OSPF (0pen Shortest Path Fir)
st) and RIP (Routing Information Protocol) are commonly used together to set routes using dynamic route control protocols. Alternatively, for all dynamic route settings, it is necessary to collect information in an optimal method for each, convert them to each other, and analyze them in a unified manner. For example, it is necessary to acquire information indicating a link state for OSPF, and to check up / down information for an adjacent route control device and from which adjacent device the route is learned in RIP. Therefore, there is a need for a method capable of promptly detecting a failure in the network and optimizing the route even if the failure occurs in the network.

As described above, there are many problems in applying the conventional load balancing method to the dynamic TE, and it is difficult to dynamically optimize the route in the network at a practical level. . The present invention is intended to solve such a problem, and an object of the present invention is to provide a route control method and apparatus capable of dynamically optimizing distribution of traffic over a plurality of routes in a network.

[0015]

In order to achieve such an object, a routing control method according to the present invention is connected to a packet transfer network and determines the flow to which the input packet belongs from the attribute of the input packet. In a routing control method used by a routing control device for identifying and outputting a packet to any route based on that flow, the flow of an input packet is determined based on a preset weighting factor of each group. The steps of classifying the flow of each group to a different route for each group, measuring the bandwidth used for each route for each group, and measuring the ratio of each measured bandwidth. Adjusting the weighting coefficient of each group based on the above.

Further, there is provided a step of calculating a unit use bandwidth for each group by dividing each measured bandwidth by a weight coefficient previously assigned to the group, and when adjusting the weight coefficient, You may make it adjust the weighting factor of each group based on unit use bandwidth.

When adjusting the weighting factors, each weighting factor may be increased or decreased so that the unit use bandwidth of each group is averaged. In addition, if the unit usage bandwidth of the group is larger than the unit usage bandwidth of the other group, the weighting coefficient of the group is increased, and if it is smaller than the unit usage bandwidth of the other group, the weight of the group is The coefficient may be reduced. Also,
When adjusting the weighting factors, the weighting factors may be adjusted so that the ratio of the weighting factors of each group becomes equal to the ratio of each bandwidth.

When classifying flows, the number of flows classified into each group may be classified so as to be proportional to each weighting coefficient. In addition, as each flow,
A connection of the OSI reference model layer 4 may be used, or a series of input packets in which at least one of the source and destination addresses of the OSI reference model layer 3 is the same may have the same flow.

Further, when classifying the flows, a hash value of the input packet is calculated by a hash function including at least one of the source address, the destination address, the source port number, and the destination port number of the input packet as an argument, It is possible to classify input packets with the same obtained hash value into the same group, or use a flow storage table that stores each flow of the classified input packets and store a new input packet in the flow storage table. If the input packet belongs to any of the flows that are registered in the flow storage table, the input packet is classified into the same group as that flow. Even if the new flow is classified into a group in which the weighting coefficient of There.

As the route, a route distinguished based on the MPLS LSP may be used. In addition, P
Virtual connection of PP, virtual connection of ATM V
It is also possible to use different routes based on C or the virtual path VP, a VLAN defined in IEEE802.1Q, or a node of the next hop of the route control device.

Further, the route control device according to the present invention is connected to the packet transfer network, identifies the flow to which the packet belongs from the attribute of the input packet, and determines the route of the packet by using the flow as a unit. In the route control device, the flow control unit that outputs the flow of the input packet to one of the groups based on the preset weighting factor of each group, and the flow of each group for each group Based on the ratio of the bandwidth output part that outputs to different routes, the bandwidth measurement part that measures the bandwidth used for each route for each group, and the bandwidth of each group measured by this bandwidth measurement part. And a weighting factor adjusting unit for adjusting the weighting factor of each group.

Further, there is provided a unit bandwidth calculating unit for calculating a unit use bandwidth for each group by dividing each bandwidth measured by the bandwidth measuring unit by a weighting coefficient previously assigned to the group. The weighting factor adjusting unit may adjust the weighting factor of each group based on each unit used bandwidth calculated by the unit bandwidth calculating unit.

When adjusting the weighting factors, the weighting factor adjusting unit may increase or decrease each of the weighting factors so that the unit use bandwidth of each group is averaged. In addition, if the unit usage bandwidth of the group is larger than the unit usage bandwidth of the other group, the weighting coefficient of the group is increased, and if it is smaller than the unit usage bandwidth of the other group, the weight of the group is The coefficient may be reduced. Further, when adjusting the weighting factors, the weighting factors may be adjusted so that the ratio of the weighting factors of each group becomes equal to the ratio of each bandwidth.

In the flow classification unit, when classifying flows,
You may make it classify so that the number of the flow classified into each group may be proportional to each weighting coefficient. In addition, the connection of the OSI reference model layer 4 may be used as each flow, or a series of input packets in which at least one of the source and destination addresses of the OSI reference model layer 3 is the same. May be set.

When the flow classifying unit classifies flows, the hash value calculating unit uses the hash function including at least one of the source address, the destination address, the source port number, and the destination port number of the input packet as an argument. The hash value of the input packet may be calculated according to, and the hash value group classification unit may classify the input packets having the same hash value obtained by the hash value calculation unit into the same group.

In the flow classification unit, when classifying flows,
A flow storage table that stores each flow of classified input packets is provided, and if a new input packet belongs to any flow stored in the flow storage table, the input packet is classified into the same group as that flow. If the input packet belongs to a new flow that is not stored in the flow storage table, the new flow may be classified into a group in which the weighting coefficient of the group should be increased.

As the route, a route distinguished based on the MPLS LSP may be used. Besides this, PP
P virtual connection, ATM virtual connection VC
Alternatively, a virtual path VP, a VLAN defined by IEEE802.1Q, or a route distinguished based on the node of the next hop of the route control device may be used.

[0028]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a schematic configuration diagram of a packet transfer network to which a route control method according to an embodiment of the present invention is applied. The packet transfer network N11 is provided with one load balancer B11 and three routers R11, R12, R13 having a route control device using the route control method according to the embodiment of the present invention. They are mutually connected by L11. The routers R11, R12, and R13 are all devices that transfer packets, and switches, hubs, repeaters, packet switches, and the like can be used as these routers. Although an example of using three routers is shown below, any number of routers can be used.

The load balancer B11 includes a transmission means L11.
A plurality of terminals T11 and T12 are connected via.
The case where one load balancer B11 is used is described below as an example, but an arbitrary number of load balancers can be used. In addition, the load balancer B11 and the terminals T11 and T12
Are directly connected by the transmission means L11, but an arbitrary number of routers and networks may exist between them. Furthermore, the load balancer B11 need not be provided at the entrance of the network as in FIG.
It is also possible to install it in 11, T12 or in routers R11, R12.

The transmission means L11 need not all be the same, and different techniques can be used for the transmission means for connecting T11-B11 and the transmission means for connecting B11-R11, for example. In addition, although not shown in the figure, a very large number of terminals may be connected to the packet transfer network N11 via other load balancers and routers, and flow through each part in the network N11. The amount of traffic may change from moment to moment.

FIG. 2 is a conceptual diagram showing a plurality of routes when a packet is transferred from the terminal T11 or the terminal T12 to the terminal T13 via the load balancer B11. In FIG. 1 described above, a plurality of routes from the load balancer B11 to the terminal T13 via the routers R11, R12, and R13 can be considered. Here, in order to facilitate understanding, consider a case where two routes P11 and P12 are used. As the route P11, a route of load balancer B11 → router R11 → router R13 → terminal T13 is used, and as route P12, a route of load balancer B11 → router R12 → router R13 → terminal T13 is used.

The packets output from the terminals T11 and T12 reach the terminal T13 through either the route P11 or the route P12 while repeating the demultiplexing / concentration with the traffic from other terminals. Each router R11 on the way,
Since the loads of R12 and the transmission means L11 are changing from moment to moment, the target of the load balancer B11 is to set these routes P11 and P12 so that the utilization efficiency of the network resources becomes highest according to the load of each route. It is to transfer traffic.

FIG. 3 is a block diagram showing a configuration example of the load balancer B11. The load balancer B11 is provided with a flow classification unit 1, flow output units 2 and 3, a bandwidth measurement unit 4, a unit bandwidth calculation unit 5, and a weight coefficient adjustment unit 6. In the load balancer B11, each packet having the same attribute is input based on the attribute of the input packet, for example, the information using the source address, the destination address, the source port number, the destination port number, etc. alone or in combination. Are managed as one flow. In addition, these flows are classified into a plurality of groups, and the bandwidth of each route used in these groups is adjusted, whereby the load on each route is distributed and controlled.

The flow classification unit 1 is a functional unit that classifies the flow to which an input packet belongs into each group based on the weighting coefficient preset for each group. The flow output units 2 and 3 are functional units which are provided for the respective routes P11 and P12 and which output the flows classified into the respective groups by the flow classification unit 1 to the relevant routes in units of the groups.
The bandwidth measurement unit 4 is a functional unit that measures the bandwidth used by the group on each path.

The unit bandwidth calculation unit 5 divides the used bandwidth of each group measured by the bandwidth measurement unit 4 by the weighting coefficient of the group to obtain the unit used bandwidth per unit weighting coefficient of each group. It is a functional unit for calculating. The weighting factor adjusting unit 6 is a functional unit that adjusts the weighting factor of each group so that the unit use bandwidth of each group calculated by the unit bandwidth calculating unit 5 is averaged. Each of these functional units may be configured by individually cooperating hardware such as a microprocessor such as a CPU and its peripheral circuits and software executed by the microprocessor, or by configuring only the hardware circuit. May be.

In the following, each flow is divided into two groups G1.
1 and G12, the flow of the group G11 is output from the flow output unit 2 to the route P11, and the flow of the group G12 is output from the flow output unit 3 to the route P12 as an example.

FIG. 4 is a flow chart showing a route control process in the load balancer. The packets input from the terminals T11 and T12 are classified into the groups G11 and G12 based on the weighting coefficient of each group in the flow classification unit 1 (step 100). As a classification method, for example, the number of flows classified into each group is classified so as to be proportional to the weighting coefficient. In FIG. 3, the number of input flows is 6, and the weighting factors W11 and W of the groups G11 and G12 are
The case where 12 is 1 and 2 is shown as an example.
According to the above classification method, two of the input flows are classified into the group G11 and the other four are classified into the group G12.

The groups G11 and G12 are output to the routes P11 and P12 by the flow output units 2 and 3, respectively (step 101). With respect to the flow of the packets output to each of the routes P11 and P12, the bandwidth measurement unit 4 sequentially measures the used bandwidth for each group (step 1).
02). Based on the measurement result, the unit bandwidth calculator 5 calculates the bandwidth per unit weighting factor (step 103). For example, the measured paths P11, P12
If the bandwidths used at 10 and 12 are respectively divided by the corresponding weighting factors 1 and 2,
The unit use bandwidth per unit weight coefficient is 10 and 6 respectively.
become.

The weighting factor adjusting unit 6 adjusts the weighting factor of each group based on the unit use bandwidth for each group calculated by the unit bandwidth calculating unit 5 (step 10).
4). Therefore, thereafter, when re-classifying each flow in the flow classifying unit 1 or classifying a new flow, the process returns to step 100, and the weight coefficient adjusted by the weight coefficient adjusting unit 6 is used. Since the bandwidth used by the flows through the paths is averaged, the distribution of traffic across multiple paths in the network will be dynamically optimized. As a result, it is possible to improve the utilization efficiency of network resources for highly bursty traffic and to provide a large number of customers with a low-cost packet transfer function.

The present invention has one aspect of providing a packet transfer method in which the transfer route is dynamically changed according to the load condition of the network. A specific description will be given taking as an example the case where packets belonging to the same TCP connection are regarded as flows. TCs included in each group
Traffic is classified so that the ratio of the number of P connections is equal to the ratio of the weighting factors assigned to each group, and the traffic is sent to different routes, and the used bandwidth of the group sent to each route is measured for each group. Then, by dividing the measured used bandwidth by the weight coefficient of the group, the unit used bandwidth per unit weight coefficient is calculated for each group, and the unit used bandwidth per unit weight coefficient becomes even among the groups. Thus, the weighting coefficient is optimized. As a result, even if there is a link or shaper that has used up the physical bandwidth to the full extent on the route, it is possible to quantitatively compare and optimize the throughput of TCP sessions passing through each route. It is possible to solve the first problem.

Further, the above-mentioned measurement and calculation of optimization can be independently performed by each load balancer without using information on loads and failures from other load balancers and routers. There is no need to install the server, which is the second issue. In addition, since the bandwidth used by each group described above can be internally measured and processed by each load balancer only for the flow passing through the device, information about all pass points of all routes over the entire network is provided. Collection is unnecessary. Therefore, the number of measurement targets is smaller than that in the conventional method, and the measurement process is further distributed to the plurality of load balancers, so that the third problem described above can be solved.

Further, the above-mentioned fourth problem can be solved by the above-mentioned method. This utilizes the fact that a transport layer protocol such as TCP generally has a congestion control function. For example, a terminal that communicates using TCP dynamically increases or decreases the bandwidth used by the connection according to the quality of the network. That is, each TCP connection automatically lowers the transmission speed when a failure or congestion occurs in the communication path, and conversely increases the transmission speed when the communication path is vacant.

In the present invention, the used bandwidth of each path is not measured as in the conventional case, but the used bandwidth per unit weighting coefficient, that is, the unit used bandwidth is calculated.
It is proportional to the bandwidth used per CP connection. Therefore, by comparing this unit used bandwidth between a plurality of routes, if the unit used bandwidth is extremely small, it is considered that some kind of failure has occurred, and if the unit used bandwidth is increasing, You can see that the route is free. As a result, it is possible to understand whether the decrease in traffic on a certain path is due to a link failure or simply because it is free, and it is possible to separate the traffic in a short time.

In the above description, the bandwidth measuring unit 4 measures the used bandwidth based on the outputs from the flow output units 2 and 3. However, the groups G11 and G12 have the routes P11, P12 and 1 respectively. Since they correspond to each other, the used bandwidth can be measured by the output from the flow classification unit 1. Further, in FIG. 4, the route control process is described as a series of flow processes, but each step in the route control process may be executed independently. In particular, the measurement interval in the bandwidth measuring section 4 can be periodically performed or can be intentionally varied. Alternatively, it may be performed as needed.

Next, referring to FIG. 5, the weighting factor adjusting unit 6
A method of adjusting the weighting factor in step S1 will be described. FIG. 5 is an explanatory diagram showing an example of adjusting the weighting coefficient. Group G1
Assuming that the weighting coefficients of 1 and G12 are W11 and W12 and the unit use bandwidth per unit weighting coefficient is BW11 and BW12, the weighting coefficient adjusting unit 6 uses the unit use bandwidth BW1.
1, the weighting factors W11 and W12 are increased / decreased according to the magnitude of BW12. That is, when BW11> BW12, W11 is increased and W12 is decreased, and BW11 <
In the case of BW12, W11 is decreased and W12 is increased. If BW11 and BW12 are the same, W11,
It is not necessary to change the value of W12.

The weighting factors W11 and W12 are assigned to each group G.
When adjusted so as to be proportional to the number of flows included in G11 and G12, the number of flows passing through the route P11 increases and the number of flows passing through the route P12 decreases as a result. Therefore, the system changes so that the bandwidth per flow becomes uniform. As for the method of adjusting the weighting factors, the degree of adjustment of the weighting factors W11 and W12 can be changed depending on the magnitude of the absolute value or the difference between the unit use bandwidths BW11 and BW12.

In the above, the concept of unit used bandwidth per unit weighting factor for each path was introduced. But,
In the step 104 (see FIG. 4) of adjusting the weighting factor described above, when the unit use bandwidth of each path is controlled to be equal, the unit use bandwidth per unit weighting factor is calculated. A configuration in which step 103 is omitted is possible. In this case, in step 104, the used bandwidth of each route measured in step 102 is used as an input variable, and each weighting factor is set so that the ratio of the weighting factor for each route becomes equal to the ratio of the used bandwidth of each route. Adjust it.

When using such a configuration,
Similarly, the bandwidth calculation unit 5 in FIG. 3 can be omitted,
At this time, the weighting factor adjusting unit 6 receives the used bandwidth of each route measured by the bandwidth measuring unit 4, and outputs a weighting factor equal to the ratio of the used bandwidth of each route. Explaining with a specific example, for example, when the measured bandwidths on the routes P11 and P12 are 10 and 12, respectively, in order to equalize the bandwidth per connection,
Connections may be classified in a ratio of 10:12 with respect to P12.

However, there is a possibility that the bandwidth used in P11 and P12 may change from the original 10:12 or the system may become unstable due to moving the connection between the routes in this way. There is. Therefore, the adjustment of the weighting factor may be divided into a plurality of times and gradually changed in accordance with the sequentially used bandwidth. Alternatively, for the purpose of accelerating the convergence speed, the variation of the weighting factor may be intentionally increased from the above by taking into consideration the variation of the band after the weighting factor is changed.

Next, with reference to FIGS. 6 and 7, a flow classification method in the flow classification unit 1 will be described. Figure 6,
7 shows an example of the flow classification method. In the flow classification method of FIG. 6, the fourth layer connection of the OSI reference model is regarded as a flow. Here, as an attribute for classification, as an example of the fourth layer protocol, TCP (Transmission
Control Protocol) is used. In this case TC
One connection of P corresponds to one flow. TCP has a congestion control function and a flow control function, and the bandwidth of each connection tends to be averaged under the same external conditions such as the amount of data, delay, and packet loss. Therefore, if W11 and W12 are adjusted so as to average BW11 and BW12 as described above, both routes P11 and P1
The system changes in the direction in which all the bandwidths of the TCP connection passing through 2 are averaged. This allows
The resource utilization efficiency of the entire network is improved.

In the flow classification method of FIG. 7, a series of packets in which at least one of the source and destination addresses of the third layer of the OSI reference model is the same is regarded as a flow. Here, as the attribute for classification, IP is used as an example of the third layer, and the flow is identified based on the source IP address. In this case, compared to the method of identifying a flow by the above-mentioned fourth layer protocol, the load for identifying a flow can be lightened. Because T
In order to identify the CP connection, generally the source IP
This is because four numbers of the address, the destination IP address, the source port number, and the destination port number are necessary, but in the case of FIG. 7, only the source IP address is required. In this example, the system is changed so that the bandwidth per source address becomes equal in the two routes. Where for example
When TCP is used as the fourth layer protocol and the source IP addresses use the same TCP connection on average, the bandwidths used by the terminals of the source IP addresses are equal. become.

In the method shown in FIG. 6 and FIG. 7, it is necessary to inspect and store which flow all belong to based on the attribute of the packet passing through the load balancer. However, it is essential in this embodiment that the input flow is G at a ratio proportional to the weight coefficient ratio, W11: W12.
11 and G12. Hash functions can be used for this purpose. FIG. 8 shows a configuration example of the flow classification unit 1 that uses a hash function having a source address, a destination address, a source port number, and a destination port number as arguments and an integer of 1 to 9 as a range. A hash value calculation unit 21 and a hash value group classification unit 22 are provided.

As the hash function, a function in which the number of flows included in each minute portion of the value range of the hash function is equal can be adopted. An example of such a hash function is a function of “(source address + destination address + source port number + destination port number) mod9 + 1” (however, AmodB is the remainder when A is divided by B). Shown). When the hash value calculator 21 calculates the hash function for each packet, an integer of 1 to 9 is obtained as a hash value. When the number of flows is large enough like the Internet, the number of flows that take each hash value is approximately equal. If it is desired to finely adjust the flow classification ratio, the number of divisions 9 of the above function may be increased. In the hash value group classification unit 22,
This value range is divided into the above W11: W12 ratios, which are associated with groups G11 and G12, respectively. As a result, the input flows can be classified as an arbitrary ratio as a result. The advantage of this method is that there is no need to store the flow through the load balancer.

Next, the distinction of routes will be described. In this embodiment, it is necessary to set a plurality of routes to the same destination inside the packet transfer network N11. In the conventional Internet, packets to the same destination follow the same route, so it is necessary to provide a method of distinguishing the routes in some way. An example of a method that can be used for this purpose is MP
LS (Multi Protocol Label Switching) LSP (La
bel Switched Path), PPP (Point-to-Point Protoc)
ol) virtual connection, frame relay virtual connection, ATM (Asynchronous Transfer Mode) virtual path VP (Virtual Path) and virtual channel VC (Virtua).
l Connection), IEEE 802.1Q VLAN (Vi
rtual Bridged Local Area Networks), next hop nodes (physical interfaces), wireless channels and frequencies, wavelengths in WDM (Wavelength Division Multiplexing), etc., but multiple routes can be distinguished. If so, the method is not limited to the above method.

In FIG. 9, the LSP of the MPLS network is
Shows the case of distinguishing routes. MPLS networks are generally labeled as Label Edge Routers.
Router / LER) and Label Switch Router (Label Swi
tching Router / LSR) and are connected to each other. Here, the label edge router is installed on the outer periphery of the MPLS network, receives a packet from an external network or terminal and transfers the packet into the MPLS network, and a packet received from inside the MPLS network to an external network. And has the function of transferring to the terminal. On the other hand, the label switch router is located inside the MPLS network and has a function of transferring a packet received from the label edge router or another label switch router to another label edge router or label switch router.

The LSP, which is a path through which packets are transferred in the MPLS network, is a label edge router (LE) at the entrance of the MPLS network.
R) is set as a starting point and the label edge router at the MPLS network exit is set as an end point, and is set along a series of label edge routers and label switch routers.
In packet forwarding, a label edge router usually obtains information such as the input interface and destination IP address of the packet, searches the routing control table based on this, and forwards the packet to each forwarding class (Forwarding Equivalen).
ce Class / FEC).

Next, the label value corresponding to each transfer class, the interface to which the packet is to be output, the OSI reference model layer 2 address of the label switch router of the next hop, etc. are searched, and the packet is searched based on these information. After encapsulating with the MPLS header and the second layer header, the packet is transferred to the corresponding label switch router of the next hop. In load balancing using the present invention in the MPLS network as described above, for example, MPLS
It suffices to install a load balancing device having the routing control device according to the present invention at the entrance of the S network and set a plurality of label switch paths starting from the device for the transfer class for which load balancing is desired. As a result, the packets classified into an arbitrary transfer class can be classified and transferred to the group corresponding to each label switch path based on each weight coefficient.

In the above, an example in which a plurality of label switch paths are associated with one transfer class has been described, but it is also possible to associate one label switch path with one transfer class. In this case, each group is associated with a transfer class on a one-to-one basis, and in the step of classifying an input packet into a transfer class, classification into groups based on weighting factors can be performed at the same time.

The place where the load balancer is installed is not limited to the entrance of the MPLS network. That is, the load balancer is installed in the middle of the label switch path that starts at any ingress label edge router and ends at the exit label edge router, and a plurality of label switch paths starting from this load balancer and reaching the exit label edge router are provided. By setting, it is possible to perform load distribution among the plurality of label switch paths starting from the load distribution device. As a result, when the quality of an arbitrary label switch path deteriorates in the middle of the route, it becomes possible to automatically direct the flow to another label switch path, and it is possible to keep the utilization efficiency of the network high at all times.

Further, FIG. 10 shows a case where the routes are distinguished by the virtual connection of PPP. PPP is OS
The I reference model is the second layer protocol, and its virtual connection generally connects two adjacent nodes. When an arbitrary node is the end point of a plurality of PPP connections, it is generally possible to clearly identify the packets belonging to each virtual connection. That is, when the transmission medium is different for each virtual connection, or when a plurality of virtual connections are set via the same transmission medium such as PPPoE (PPP over Ethernet (registered trademark)), the virtual connection is set. A unique set of {source MAC address, destination MAC address, session identifier} may be assigned to each.

As an example of load balancing using such a PPP virtual connection, a case where one customer connects to a plurality of service providers at the same time and connects to the network via these will be described as an example. In the following, for simplification of description, it is assumed that the customer and each service provider are connected by one PPP virtual connection. That is, it is assumed that the load balancer owned by the customer is connected to remote access servers of a plurality of service providers via different PPP virtual connections.

In the above case, one group can be assigned to each service provider, and the ratio of the corresponding weighting factors can be adjusted according to the bandwidth used toward each provider. That is, when one service provider is busy and the other service provider is free, it is possible to automatically direct the flow toward the busy service provider to the free service provider. it can. This makes it possible to efficiently transfer packets to each service provider even when the congestion status of a plurality of service providers changes moment by moment.

FIG. 11 shows the case where the ATM VP or VC is used. ATM VP or VC is
For example, it can be used to connect between routers. In the following, an example will be described in which there are two routes from the load balancer to any same destination, and routers at the hops next to each route are connected by different VPs or VCs. In the above case, in the load balancer, the traffic destined for the same destination is 2
The weighting factor for each group can be adjusted based on the result of measuring the bandwidths of each group.

Thus, for example, one VP or VC
Even if the same destination cannot be reached due to a failure in
Since the weighting factor for the other VP or VC is automatically increased, the flow can be moved from one path having a failure to the other path having no failure. Note that the same destination is not necessarily limited to the end point of the packet, and for example, points where there are a plurality of routes to the network on the way can be regarded as the same destination.

In FIG. 12, the VL of IEEE802.1Q is used.
The case where a route is distinguished using AN is shown. For example, when a packet can reach the same destination via a plurality of VLANs, each VLAN can be treated as one route. In this case, a weighting factor is assigned to each route, traffic passing through each VLAN is measured, and each weighting factor is adjusted based on this result, so that each VLAN is efficiently used to transfer traffic. It becomes possible.

FIG. 13 shows the case where the routes are distinguished based on the node of the next hop (when the interfaces are physically different). The node here refers to a general node having a function of transmitting / receiving or transferring a packet based on information of the network layer (or data link layer). The next hop is a node that the next node sees from any node on the route that transfers the packet from the source to the destination. Below, on the Internet,
The router of any service provider is IX (Internet e
Xchange) will be used as an example to connect to multiple other service providers.

When a plurality of service providers can be selected as the next hop for the packet to reach the same destination network, the route control method according to the present invention is applied to the route passing through the plurality of service providers. Can be used for load balancing. IX
If the structure of is a structure in which the router of each provider is connected to surround the core switch of the data link layer, the concrete means for identifying the route is the IP address of the node of the next hop and this. The data link layer address corresponding to can be used.

Packets addressed to the above destination network are classified into groups for these plural routes, the used bandwidth for each group is measured, and the value of the weighting factor is reflected, so that the quality of each route is sometimes changed. Even when it changes moment by moment, it is possible to always effectively use the resources of each route to transfer the traffic. For example, when the quality of the route passing through an arbitrary service provider is deteriorated, the flow passing therethrough can be immediately automatically redirected to the route passing through another service provider. On the contrary, when a route passing through an arbitrary service provider is vacant than another route, the flow can be automatically redirected from the other route to the vacant route.

When a provider in the middle of an arbitrary route fails and the destination network becomes unreachable, the congestion control function operates in the TCP connection via the route, and the throughput deteriorates rapidly. Therefore, by measuring this and updating the weighting factor, the failure can be promptly avoided. here,
What is particularly important is that the load balancing apparatus autonomously perform such a load balancing function and a failure avoidance function. That is, the route control can be performed only by the information obtained from the traffic passing through the load balancer without acquiring any route information or fault information from other routers or servers.

In the above example, the case where a single provider uses the above load balancing apparatus has been described, but a plurality of providers may use their respective load balancing apparatuses in parallel. Further, in the above example, the load balancing process for one destination has been described as an example, but it is also possible to perform the load balancing process for another destination at the same time while performing the load balancing process for any destination.

Further, the target of the load balancer for distributing the traffic is not limited to the same kind of route. For example, there are three routes that can reach the same destination, the first route uses the MPLS label switch path, the second route uses the PPP virtual connection, and the third route uses Even when the ATM VC is used to connect to another router, it is possible to distribute the load among these routes. The reason is that the route control method and device according to the present invention do not depend on the specific means of each route as long as the used bandwidth of the traffic passing through the load balancer can be measured for any plurality of routes. Is.

Next, referring to FIG. 14, the flow classification unit 1
The method of managing the flow in the above will be explained. FIG. 14 is a flowchart showing the flow management method. In the present embodiment, when the weighting factor ratio W11: W12 changes, the movement of the flow between the routes occurs. At this time, if there is a difference in delay time between the two routes, the order of the packets belonging to the same flow may be changed in the middle.
For example, when TCP or the like is used, changing the packet order within the same flow may cause performance deterioration.

Therefore, as shown in FIG. 14, the flow classifying unit 1 manages so that packets belonging to the same flow always pass through the same route even when the weighting factor changes. In this method, the flow storage table 7 is provided, and the flow to which the passing packet belongs and the route assigned to the flow are registered in association with each other. First, when a packet is input, the flow storage table 7 is referenced to search whether the flow to which the packet belongs is already registered (step 110). As a result, when the flow is already registered (step 111: YES), the packet is classified into the group corresponding to the flow (step 112). On the other hand, if the flow is not registered (step 111: NO), the packet is classified into a group whose weighting coefficient is to be increased, and the corresponding flow and group are newly registered in the flow storage table 7 (step 113). ).

FIG. 15 shows an example of the flow storage table 7 when a TCP connection is used as a flow.
The flow identifies four numbers of source address, destination address, source port number, and destination port number as identifiers,
An output group is associated with this. In this example, the flow identifier is {SIP1, DIP1, SPO
The packet RT1, DPORT1} is classified into G11, and the flow identifier is {SIP2, DIP2, SPORT
2, DPORT2} is classified as G12. When the flow identifier of the input packet matches any of these, it is classified into the corresponding output groove. If the flow identifier of the input packet does not match any of these, the packet can be classified into a group in which the weighting factor should be increased.

FIG. 16 shows an example of the flow allocation table 8 that can be used when determining to which group an unregistered flow should be registered. In this flow allocation table 8, a current weight (current value of the number of registered flows) and a target weight (target value of the number of registered flows) are associated with each group. The target weight can be calculated to be equal to the ratio of the weighting factors at that time. For example, the weighting factor ratio W1
Assuming 1: W12 is 1: 2, the flow identifier {S
When a packet of IP1, DIP1, SPORT3, DPORT1} is input, the corresponding flow identifier is not registered in the flow storage table 7 of FIG. Therefore, referring to the flow allocation table 8, it can be seen that the current weight of G12 is smaller than the target weight.

Therefore, while registering this packet in G12, a new flow identifier {SIP1, DIP
1, SPORT3, DPORT1} to group G12
Is registered in the flow storage table 7 in association with, and the number of registered flows corresponding to the group G12 in the flow allocation table 8 is increased by 1. For example, in the current Internet, the average duration of flows is several seconds, and by sequentially performing the above method, packets belonging to the same flow are not reordered, and the number of flows proportional to the weight coefficient is assigned to each route. It becomes possible to output. The flow storage table 7 may use an aging timer to delete the registration of an old flow.

[0077]

As described above, according to the present invention, the flow of an input packet is classified into one of the groups based on the preset weighting coefficient of each group, and the flow of each group is grouped according to the group. Output to different routes, measure the bandwidth used in each flow among the bandwidth used in each route for each group, and calculate the weighting factor of each group based on the ratio of these bandwidths. It was adjusted. Therefore, since the bandwidths used by the flows passing through the respective routes are averaged, it is possible to dynamically optimize the distribution of traffic over a plurality of routes in the network and efficiently use the network resources.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram showing a packet transfer system to which a route control method according to an embodiment of the present invention is applied.

FIG. 2 is a conceptual diagram showing a plurality of routes.

FIG. 3 is a block diagram illustrating a configuration example of a load balancer.

FIG. 4 is a flowchart showing a route control process in the load balancer.

FIG. 5 is an explanatory diagram illustrating an example of a weighting coefficient adjustment method.

FIG. 6 is an example of packet classification when the fourth layer connection of the OSI reference model is regarded as a flow.

FIG. 7 is an example of packet classification when the third layer address of the OSI reference model is regarded as a flow.

FIG. 8 is a configuration example of a flow classification unit that classifies packets by a hash function.

FIG. 9 is an example of packet output in the case of distinguishing routes by the LSP of the MPLS network.

FIG. 10 is an example of packet output when a path is distinguished by a virtual connection of PPP.

FIG. 11 is an example of packet output when a route is distinguished by a virtual connection of an ATM network.

FIG. 12 is an example of packet output in the case of distinguishing a route by a VLAN identifier of a VLAN network.

FIG. 13 is an example of packet output when a path is distinguished by an output physical interface.

FIG. 14 is a flowchart showing a flow management method.

FIG. 15 is a configuration example of a flow storage table.

FIG. 16 is a configuration example of a flow allocation table.

[Explanation of symbols]

N11 ... Packet transfer network, B11 ... Load balancing device (route control device), R11, R12, R13 ... Router, T11, T12, T13 ... Terminal, P11, P12 ...
Path, G11, G12 ... Group, W11, W12 ... Weighting coefficient, BW11, BW12 ... Unit used bandwidth, N12
... MPLS network, N13 ... PPP network, N14 ... ATM network, N15 ... VLAN network, N16 ... Physically separated network, 1 ... Flow classification unit, 2, 3 ... Flow output unit, 4 ...
Bandwidth measuring unit, 5 ... Unit bandwidth calculating unit, 6 ... Weighting coefficient adjusting unit, 7 ... Flow storage table, 8 ... Flow allocation table.

Claims (30)

[Claims] 1. A routing control device connected to a packet transfer network, which uses an attribute of a packet to identify a flow to which the packet belongs, and which outputs the packet to any route in units of the flow. In the route control method, a step of classifying an input packet flow into one of groups based on preset weighting factors of each group, and a step of outputting the flow of each group to a different route for each group And a step of measuring the bandwidth used in each path for each group, and a step of adjusting the weighting coefficient of each group based on a ratio of the measured bandwidths. And the route control method. 2. The route control method according to claim 1, wherein a unit use bandwidth is calculated for each group by dividing each of the measured bandwidths by a weighting coefficient pre-allocated to the group. The route control method further comprising the step of: when adjusting the weighting factor, adjusting the weighting factor of each of the groups based on each of the unit use bandwidths. 3. The route control method according to claim 2, wherein when adjusting the weighting factors, the weighting factors are increased or decreased so that the unit use bandwidths of the groups are averaged. Route control method. 4. The route control method according to claim 2, wherein when the weighting coefficient is adjusted, if the unit usage bandwidth of the group is larger than the unit usage bandwidth of another group, the weighting coefficient of the group is set. A routing control method comprising increasing the number of bits and increasing the weighting factor of the group when the bandwidth is smaller than the unit bandwidth of another group. 5. The route control method according to claim 1, wherein when adjusting the weighting factors, adjusting the weighting factors so that the ratio of the weighting factors of the respective groups is equal to the ratio of the respective bandwidths. A route control method characterized by: 6. The route control method according to claim 1, wherein when classifying the flows, the number of flows classified into each of the groups is classified so as to be proportional to each weight coefficient. Control method. 7. The route control method according to claim 1, wherein when classifying the flows, the OSIs are set as the respective flows.
A route control method using a connection of a reference model fourth layer. 8. The routing control method according to claim 1, wherein when classifying the flows, a series of input packets in which at least one of a source address and a destination address of an OSI reference model layer 3 is the same is the same flow. A route control method characterized by the above. 9. The route control method according to claim 1, wherein when classifying the flows, a hash including at least one of a source address, a destination address, a source port number, and a destination port number of an input packet as an argument. A route control method, wherein a hash value of the input packet is calculated by a function, and the input packets having the same hash value are classified into the same group. 10. The routing control method according to claim 1, wherein when classifying the flows, a flow storage table that stores each flow of the classified input packets is used, and new input packets are stored in the flow storage table. If the input packet belongs to a new flow that is not stored in the flow storage table, then the input packet is classified into the same group as that flow. A route control method comprising classifying the new flow into a group in which a weighting coefficient of the group should be increased. 11. The route control method according to claim 1, wherein a route distinguished based on an MPLS LSP is used as the route. 12. The route control method according to claim 1, wherein a route distinguished based on a virtual connection of PPP is used as the route. 13. The route control method according to claim 1, wherein a route distinguished based on a virtual connection VC or a virtual path VP of ATM is used as the route. 14. The route control method according to claim 1, wherein the route is a V specified by IEEE802.1Q.
A route control method using a route distinguished based on a LAN. 15. The route control method according to claim 1, wherein a route distinguished based on a node of a next hop of the route control device is used as the route. 16. A routing control device that is connected to a packet transfer network, identifies the flow to which the inputted packet belongs from the attribute of the inputted packet, and outputs the packet to any of the routes on the basis of the flow. In the device, a flow classifying unit that classifies the flow of the input packet into one of the groups based on a preset weighting coefficient of each group, and a flow that outputs the flow of each group to a different route for each group. An output unit, a bandwidth measurement unit that measures the bandwidth used in each path for each group, and a bandwidth measurement unit for each group based on the ratio of the bandwidth of each group measured by the bandwidth measurement unit. A route control device, comprising: a weighting factor adjusting unit that adjusts the weighting factor. 17. The route control device according to claim 16, wherein each of the bandwidths measured by the bandwidth measuring unit is divided by a weighting factor pre-allocated to the group, and a unit is obtained for each group. A unit bandwidth calculation unit that calculates a used bandwidth, wherein the weight coefficient adjustment unit adjusts the weight coefficient of each group based on each unit used bandwidth calculated by the unit bandwidth calculation unit. A path control device characterized by: 18. The route control device according to claim 17, wherein the weighting factor adjusting unit increases or decreases each weighting factor so that the unit use bandwidths of the groups are averaged. Control device. 19. The route control device according to claim 17, wherein the weighting factor adjusting unit increases the weighting factor of the group when the unit usage bandwidth of the group is larger than the unit usage bandwidth of another group. A routing control device that reduces the weighting coefficient of the group if the unit usage bandwidth of the other group is smaller than that. 20. The route control device according to claim 16, wherein the weighting factor adjusting unit adjusts the weighting factor such that a ratio of weighting factors of each group is equal to a ratio of each bandwidth. A characteristic route control device. 21. The route control apparatus according to claim 16, wherein the flow classification unit classifies the number of flows classified into each group so as to be proportional to each weighting coefficient. apparatus. 22. The route control device according to claim 16, wherein the flow classification unit uses a connection of an OSI reference model layer 4 as each of the flows. 23. The routing control device according to claim 16, wherein the flow classification unit sets a series of input packets in which at least one of a source address and a destination address of the third layer of the OSI reference model is the same as the same flow. A path control device characterized by: 24. The routing control device according to claim 16, wherein the flow classifying unit has at least one of a source address, a destination address, a source port number, and a destination port number of the input packet as an argument. And a hash value group classifying unit for classifying input packets having the same hash value obtained by the hash value calculating unit into the same group. Route control device. 25. The routing control device according to claim 16, wherein the flow classification unit has a flow storage table that stores each flow of the classified input packets, and a new input packet is stored in the flow storage table. If the input packet belongs to a new flow that is not stored in the flow storage table, then the input packet is classified into the same group as that flow. A path control device which classifies the new flow into a group in which a weighting coefficient of the group should be increased. 26. The route control device according to claim 16, wherein a route distinguished based on an LSP of MPLS is used as the route. 27. The route control device according to claim 16, wherein a route distinguished based on a virtual connection of PPP is used as the route. 28. The route control device according to claim 16, wherein a route distinguished based on a virtual connection VC or a virtual path VP of ATM is used as the route. 29. The route control device according to claim 16, wherein the route is a V specified by IEEE802.1Q.
A route control device characterized by using a route distinguished based on a LAN. 30. The route control device according to claim 16, wherein when the route is identified, a route distinguished based on a node of a next hop of the route control device is used.