JP2002118585A - Method and apparatus for scheduling packet - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method and an apparatus for scheduling packet in which the quality of an application requiring a strict real time performance is prevented from deteriorating by utilizing all available bands effectively, and transmission delay time is prevented from increasing even for an application not requiring a strict real time performance. SOLUTION: Input packets are sectioned into a plurality of classes which are then sorted into a preferential group and a nonpreferential group, and packets in each class are stored in queues independent from each other. First band satisfying a requested delay time is assigned preferentially to a queue holding the packets of the preferential group, and remaining bands are determined as a second band. Use frequency of a specific interval is detected for each class of the nonpreferential group, and a part of the second band is assigned dynamically to the queue of nonpreferential group at a weight corresponding to the use frequency of a nearest specific interval.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a packet scheduling method and a packet scheduling apparatus. For example, various routers in an autonomous system (AS) provided in an IP (Internet Protocol) network determine a processing order of input packets. Used to

[0002]

2. Description of the Related Art For example, various types of packets are sequentially input to an input of a router provided in an IP network or the like. A plurality of queues (queues) are provided inside a router that processes various types of packets, and input packets are stored in queues corresponding to the types and then processed.

[0003] The process of determining the order in which packets stored in a plurality of queues are processed is called packet scheduling. The following techniques are conventionally known as packet scheduling methods. (1) Round Robin: A packet in the first queue, a packet in the second queue, a packet in the third queue,... Are sequentially processed. The number of packets to be processed at one time is determined in advance (generally, 1
One). Processing is skipped for queues in which there are no packets to be processed.

(2) WRR (Weighted Round Robin):
When the “weight” and the packet length are different for each packet class (queue), the “normalized weight” obtained by dividing the weight by the average packet length is used, and the number of packets corresponding to the weight is determined by the round robin. To process. (3) DRR (Deficit Round Robin): In the WRR method, it has been devised so that information on the average packet length is not required in advance. The details are processed in the following procedure. There is a certain "reference amount" for each class (queue)
If the length of the packet at the head of a certain queue is equal to or smaller than the reference amount, the packet is processed. Then, the reference amount is reduced by the packet length. In this way, the queue is processed continuously until the reference amount is less than the length of the packet to be processed next. When it is no longer satisfied, the processing target moves to the next queue in a round robin manner. When the head packet length of all queues becomes larger than the reference amount of each queue, a new reference amount is added to all queues.

(4) GPS (Generalized Processor Sh)
aring): This is a scheduling method in which servers having an infinite processing capacity simultaneously process N / queues where packets exist at a service rate of 1 / N. If there is no packet to be processed in one queue, the processing capacity is equally allocated to other queues. This is an ideal method that cannot be realized in practice.

(5) WFQ (Weighted Fair Queuein)
g): This is a method for approximately realizing the GPS method. In the WFQ, packet processing is terminated in bit units, while the GPS performs processing in infinitesimal units at the same time. The scheduled time is calculated, and the processing is performed in order from the packet with the earliest time. (6) CB-WFQ (Class-based Weighted Fair Queu
eing): A weight value is fixedly assigned to each queue.

[0007] For example, Japanese Patent Application Laid-Open No. 9-83547 discloses that scheduling is performed based on the weight set for each packet queue and the packet accumulation amount of the packet queue. In addition, when selecting a packet scheduling method, it is general to select the method in consideration of the following items.

(A) What and how to achieve by scheduling (for example, differences in priority, fairness,
You want to end the processing within the upper limit of the delay time, etc.). (B) How to define the priority and whether it is actually realized. (C) Whether fairness between classes needs to be considered, and if so, how?

(D) How easy is it to actually implement it in the apparatus? The most convenient method is selected in consideration of a combination of the above items.

[0010]

For example, in an IP network, various types of packets are mixedly transmitted to carry information of various types of applications. Application types include voice (VoIP) and image (Vi
deo on Demand and video conferencing), Telnet, X-W
Indow, WWW, electronic mail, FTP and the like.

Packets that carry audio and video application information are called "stream-type traffic."
A packet carrying information of an application such as Telnet or X-Window is called "interactive traffic". Further, WWW, e-mail, F
A packet that carries information of an application such as a TP is called "bulk transfer type traffic".

[0012] For "stream-type traffic" and "interactive-type traffic", real-time performance is important, and if the delay time during transmission is increased, the transmission quality is remarkably deteriorated. In addition, real-time performance is not so important for “bulk transfer type traffic”. The CB-WFQ
When adopting the method, "stream type traffic"
Priority is assigned to large queues for processing traffic and "interactive traffic", and low weight is assigned to queues for processing "bulk transfer traffic" in advance, ensuring quality for applications that require real-time performance. can do.

However, when the weight of each queue is fixedly determined as in the CB-WFQ system, the quality of a specific application using a queue to which a small weight is assigned may be significantly reduced. That is, the frequency of use of a particular queue increases, and when many packets arrive within a short time compared to the bandwidth allocated to the queue, the delay time in the queue increases significantly.

When the technique disclosed in Japanese Patent Application Laid-Open No. 9-83547 is used, scheduling is dynamically performed in accordance with the amount of stored packets in a packet queue, so that the priority of packets of each class changes sequentially. Will be. Therefore, a bandwidth cannot be guaranteed for a specific class of packets having high real-time properties. Further, the size of the band allocated to a specific class may be too small.

According to the present invention, in a packet scheduling method and a packet scheduling apparatus, all available bandwidths are effectively used to prevent quality degradation of an application strict in real-time performance, and an application that requires a small real-time performance is required. It is another object of the present invention to prevent the transmission delay time from increasing.

[0016]

A first aspect of the present invention is a packet scheduling method for storing sequentially input packets as a queue and determining the order of packets to be taken out of the queue for processing. The input packets are divided into a plurality of classes, and the plurality of classes are classified into a priority group corresponding to stream-type traffic or interactive traffic, and a non-priority group corresponding to bulk-transfer-type traffic. Are stored in queues independent of each other, and a first band sufficient for satisfying the required delay time of the corresponding class is assigned to the queue holding the packets of the class belonging to the priority group. Is assigned preferentially, and the sum of the first band is subtracted from the value of the entire band. For each class belonging to the non-priority group, the use frequency in a specific period is detected for each class, and the latest band is stored in a queue holding packets of each class belonging to the non-priority group. A part or the whole of the second band is dynamically allocated with a weight according to a difference in use frequency detected in a specific period of time, and an order of packets taken out from the plurality of queues is allocated to each queue. Is determined according to the size of the band.

In the present invention, a first band sufficient to satisfy a required delay time of each class is preferentially allocated to a queue holding packets of each class belonging to the priority group. The queue holding packets of each class belonging to the non-priority group includes:
A part or the whole of the second band is dynamically allocated with a weight according to the difference in the use frequency detected in the latest specific period. The second band is a result of subtracting the sum of the first band from the value of the entire band.

Therefore, the queues for processing the stream-type traffic and the interactive-type traffic can preferentially process the packets so as to satisfy the delay time required by each application. In addition, a weight corresponding to a difference in the frequency of use of each class detected in the latest specific period is reflected in the size of the band allocated to the queue for processing the bulk transfer type traffic. That is, a part or the whole of the second band is dynamically allocated to each class of the non-priority group.

For this reason, also in the class for processing the packets of the non-priority group, when the number of packets input to the queue increases rapidly, the bandwidth allocated to the queue for processing the packets of the class increases. Therefore, an increase in delay time can be suppressed. According to a second aspect of the present invention, in the packet scheduling method according to the first aspect, when the non-priority group includes a specific class whose allocated bandwidth is too small, priority is given to a class belonging to the priority group. First, a third band is first allocated to a queue holding packets of the specific class, and then the first band is allocated to a queue holding packets of a class belonging to the priority group. It is characterized by assigning.

If only some classes of packets are processed preferentially, the bandwidth allocated to a specific class becomes very small, and there is a possibility that packets of a specific class will not be processed at all. In claim 2, when the size of the band allocated to the specific class is considered to be too small, the third band is preferentially allocated to the specific class. It can be processed reliably. That is, the minimum bandwidth can be guaranteed for the class of the non-priority group.

According to a third aspect of the present invention, in the packet scheduling method according to the first aspect, the frequency of use of each class of the non-priority group is detected based on the number of packets arriving in the specific period and the average packet length of the arriving packets. It is characterized by doing. According to claim 3, the frequency of use of each class is detected based on the number of packets arriving during the specific period and the average packet length of the arriving packets, so that the length of each arriving packet varies. Can also detect the frequency of use of each class.

A fourth aspect of the present invention is a packet scheduling apparatus for storing sequentially input packets as a queue and determining the order of packets to be taken out of the queue for processing. Classifying means into a plurality of classes, classifying means for classifying the plurality of classes into a priority group corresponding to stream type traffic, and a non-priority group corresponding to bulk transfer type traffic; Priority is given to the first band sufficient to satisfy the required delay time of the corresponding class with respect to the storage control means for storing the packets belonging to the priority group and the queue for storing the packets belonging to the class belonging to the priority group. Subtracting the sum of the first band from the value of the entire band. Band calculation means for determining the result as the second band, use frequency detection means for detecting use frequency in a specific period for each class belonging to the non-priority group, and packets of each class belonging to the non-priority group A second band allocating means for dynamically allocating a part or the entirety of the second band with a weight corresponding to a difference in use frequency detected in a latest specific period, Scheduling means for determining the order of packets to be taken out from the plurality of queues according to the size of the bandwidth allocated to each queue.

According to the fourth aspect of the present invention, packet scheduling can be performed in the same manner as in the first aspect.

[0024]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of a packet scheduling method and a packet scheduling apparatus according to the present invention will be described with reference to FIGS. This form corresponds to all claims.

FIG. 1 is a flowchart showing a packet scheduling process of this embodiment. FIG. 2 is a block diagram showing a configuration of a router for implementing the present invention. FIG. 3 is a time chart showing the timing of the processing of this embodiment. In this embodiment, the classifying unit, the storage control unit, the first band allocating unit, the band calculating unit, the use frequency detecting unit, the second band allocating unit, and the scheduling unit according to claim 4 respectively include the class determining unit 11, the packet information Collection unit 10, S22, S23, traffic characteristic amount calculation unit 1
6 (S15), S24, and the processing class instruction unit 18.

In this embodiment, it is assumed that the packet scheduling method and the packet scheduling device of the present invention are applied to the router of FIG. 2 connected to the IP network. Referring to FIG. 2, the router includes a packet information collecting unit 10, a class determining unit 11, a header reading and measuring unit 12, a packet header analyzing / calculating unit 13, a routing table 14, a header writing control unit 15, a traffic characteristic amount calculating unit. 16, a bandwidth calculation unit 17, a processing class instruction unit 18, a database (DB) 19, a scheduling unit 20, a clock 31, and a timer 32.

The scheduling unit 20 includes N queue buffers 21 (1) to 21 (N) and a packet processing unit 22. The queue buffer 21 is, for example, F
It can be constituted by an IFO (first in first out) memory. A packet arriving at the router from the outside is input to the packet information collecting unit 10, and the scheduling unit 20 selects one of the queue buffers 21.
After that, it is processed by the packet processing unit 22 and output to the outside.

The packet information collecting unit 10 extracts header information including the destination, class, and packet length from each packet, and notifies the class determining unit 11 of the extracted header information. The class determining unit 11 classifies the types of packets into N or less classes based on the class of the header information. Further, a class of the priority group and a class of the non-priority group are distinguished. The packet information collecting unit 10 writes the input packet into one queue buffer 21 corresponding to the class classified by the class determining unit 11.

The header information including the destination and the class of each packet is input from the class determination section 11 to the packet header analysis / calculation section 13 through the header reading / measuring section 12. The packet header analysis / calculation unit 13 acquires destination information by referring to the routing table 14 based on the header information. This destination information passes through the header write control unit 15 and is written in the packet on the queue buffer 21 as next router information.

The traffic characteristic amount calculation section 16 calculates a traffic characteristic amount for each specific section for each packet class. The allocation bandwidth calculation unit 17 allocates the bandwidth of each class according to the characteristic amount obtained by the traffic characteristic amount calculation unit 16. The value of the band allocated by the allocated band calculation unit 17 is input to the processing class designating unit 18 as the weight of each class. Further, the number of packets to be processed in each specific section is input to the processing class instructing unit 18 from the traffic characteristic amount calculating unit 16.

The processing class designating unit 18 indicates the class of the packet to be processed by the packet processing unit 22 to the packet processing unit 2.
Instruct 2 The packet processing unit 22 extracts a packet of the class specified by the processing class instruction unit 18 from the head of the corresponding queue buffer 21 and processes the packet. The packet processing unit 22 in FIG. 2 uses queue buffers 21 (1) to 21 (N).
The details of the packet scheduling process for determining the order of the packets to be extracted from the packet will be described below.

The class of the packet may be determined depending on the type of the application.
It may be determined at the time of a user contract or may be determined at the start of communication. Application types include voice (VoIP), images (Video on Demand and video conference), Telnet, X-Window, and WW
W, e-mail, FTP, etc.

Packets that carry information about audio and video applications are called "stream-type traffic."
A packet carrying information of an application such as Telnet or X-Window is called "interactive traffic". Further, WWW, e-mail, F
A packet that carries information of an application such as a TP is called "bulk transfer type traffic".

For "stream-type traffic" and "interactive-type traffic", real-time performance is important, and if the delay time during transmission increases, the transmission quality significantly decreases. In addition, real-time performance is not so important for “bulk transfer type traffic”. Therefore, in this packet scheduling process, the class of the packet is classified into a priority group and a non-priority group. Packet classes corresponding to “stream type traffic” and “interactive type traffic” are classified into priority groups because real-time properties are important. The class of the packet corresponding to the “bulk transfer type traffic” is classified into the non-priority group because the real-time property is not important. Each class belonging to the priority group is called a priority class, and each class belonging to the non-priority group is called a non-priority class.

The packet scheduling process shown in FIG. 1 represents a procedure for allocating a bandwidth to a queue (contents of each queue buffer 21) for processing packets of each class. The priority of processing packets stored in each queue is determined according to the size of the bandwidth allocated to each queue. Further, the packet scheduling process shown in FIG. 1 is executed at a timing as shown in FIG. That is, if the current time is t, information is collected for the information collection section A1 before that, and the calculation section A
The calculation is performed in step 2, and the band is allocated to the control target section A3.

Now, when a packet arrives at the input of the router of FIG.
The process proceeds from step S12 to S13. In step S13, the class of the arrived packet is identified. In the next step S14, the arriving packet is transferred to one queue buffer 21 associated with the class. Further, in step S15, the number of packets arriving for each class and the average packet length are calculated.

At a predetermined timing for allocating a band to each class, the process proceeds from step S11 to S16. Steps S16 to S19 are processing for guaranteeing a minimum bandwidth for non-priority class packets.
In step S16, it is determined whether or not there is a non-priority class in a certain monitoring section (between t-KT and t) for which the allocated bandwidth is too small to process any packet. If it exists, the process proceeds to step S18; otherwise, the process proceeds to step S17.

In step S17, the band BWα (bp
Set the value of s) to 0. In step S18, the result of multiplying the total bandwidth BWt (bps) of the router by the coefficient X (for example, 0.1) is set as the value of the bandwidth BWα. The coefficients X and K are obtained by calculation so that the delay time of each class does not exceed a certain value. In step S19, the bandwidth BWα is allocated to the non-priority class that has failed to process the packet.

When there are a plurality of non-priority classes that cannot process a packet, a part of the band BWα is allocated to each non-priority class so that the delay time of each class does not exceed a certain value. In step S20, the result of subtracting the band BWα from the entire band BWt is determined as the value of the band BWβ.

In step S21, the stay time Tisys (A1) of the packet observed in the information collection section A1 in the node (router) is calculated for each class (only the priority class) according to the following equation. Tisys (A1) = xi (A1) + pi (A1) xi (A1) / (2 (1-pi (A1))) (1) where xi (A1) = 8pi (A1) / BW (Ci , A1): average processing time (sec) pi (A1): average packet length (bytes) BW (Ci, A1): size of band allocated to the ith class ρi (A1) = λi (A1) xi ( A1): Usage rate λi (A1) = qi (A1) / nT: Arrival rate qi (A1): Number of packets of the i-th class arriving in the information collection section A1 In step S22, the bandwidth of the priority class is determined for each class. Update assignments. Specifically, step S21
Band B such that the stay time Tisys (A1) of the packet obtained in the above becomes smaller than the delay time di required by each class.
The size of W (Ci, A3) is assigned. That is, for the priority class, a band satisfying the condition of the delay time di is given to the control target section A3 based on information obtained in the section (A1) of the latest n cycles.

Accordingly, the bandwidth is allocated to all the priority classes so as to substantially satisfy the delay time di required by the priority classes. Also, for a packet of a priority class that cannot satisfy the delay time di, the call reception is controlled in advance to reject such a call reception. In step S23, first, the total sum BWr of the bandwidths BW (Ci, A3) allocated to all the classes belonging to the priority group in step S22 is obtained. Next, a result obtained by subtracting the band BWr from the band BWβ is obtained as a remaining band BWz.

In step S24, the remaining bandwidth BWz is assigned to each non-priority class. Specifically, based on the number N (Ck, A1) of arriving packets and the average packet length PL (Ck, A1) (only the non-priority class) obtained in the information collection section A1, each k-th class The band BW (Ck, A3) of Ck is obtained using the following equation. BW (Ck, A3) = BW
z · uk · vk / (sum of uk · vk of non-priority class)
(2) However, uk = N (Ck, A1) (only the non-priority class) vk = PL (Ck, A1) (only the non-priority class) That is, for the packet of the non-priority class, the packet arriving in the information collection section A1 The bandwidth BWz is allocated to each class at a rate proportional to the information amount (the number of packets × the average packet length). For this reason, as the frequency of use of the non-priority class increases, more and more bandwidth is allocated. Also, as the frequency of use decreases, the allocated bandwidth decreases.

The size of the bandwidth BW (Cj) (j = 1 to N) assigned to each class by the processing of FIG. 1 corresponds to the "weight" at the time of router scheduling, and the priority in the control target section A3 Directly connected to That is, the priority weight W (Cj, Cj,
A3) is given by the following equation.

W (Cj, A3) = BW (Cj, A3) / BWt (3) However, for each class of the non-priority group, only when there is no packet in the queue of each class of the priority group , A queue is selected according to the above weights, and the packet is processed. Next, a specific example of band allocation will be described.

(Specific Example 1) In this example, the following state is assumed. BWt: 12 Mbps BWα: 0 T: 5 seconds n: 1 m: 1 class number N: 3 Class 1: Priority class Class 2, Class 3: Non-priority class At present t: 0 Also, information collection section A1 (−5, 0 The following state is assumed as the use state of each class in ()).

Number of arriving packets of class 1: 2 Number of arriving packets of class 2: 2 Number of arriving packets of class 3: 6 Average packet length of class 1: 100 Average packet length of class 2: 200 Average packet length of class 3: Here, assuming that the bandwidth BWr is 8 Mbps, the bandwidth allocated to each class for the control target section A3 (5, 10) is as follows.

Class 1: 8 Mbps Class 2: (12-8) × 2 × 200 / (2 × 200 + 6 × 300) = 0.727
Mbps class 3: (12-8) x 6 x 300 / (2 x 200 + 6 x 300) = 3.273
Mbps (Specific Example 2) In this example, an integral multiple of the observation period T (KT)
In a certain section, a case where the number of packets of the priority class arrived too much, the bandwidth allocation to the non-priority group was reduced, and a packet of a certain class (unprocessed class) belonging to the non-priority group was not processed at all. Suppose.

In this case, for each section of (KT), the band of X (%) of the entire band in the next control target section A3 is assigned to the unprocessed class and processed. The coefficients K and X are
The calculation is made so that the delay time of each class does not exceed a certain value. That is, when an unprocessed non-priority class is detected, the bandwidth BWα is determined in step S19 of FIG.
To the outstanding non-priority class.

The following state is assumed here. BWt: 12 Mbps T: 5 seconds n: 1 m: 1 class number N: 3 class 1: priority class class 2: non-priority class class 3: non-priority class (unprocessed class) K: 2 X: 10 (%) t: 0 Further, the following state is assumed as the use state of each class in the information collection section A1 (−5, 0).

Class 1 arriving packets: 9 Class 2 arriving packets: 1 Class 3 arriving packets: 0 Class 1 average packet length: 100 Class 2 average packet length: 200 Class 3 average packet length: 300 Here, the bandwidth BWr is 11 Mbps, and the section (−1)
It is assumed that no class 3 packet is processed in (0,0). In this case, the control target section A3 (5,
The bandwidth allocated to each class for 10) is as follows.

Class 1: 12−1.2 = 10.8 Mbps Class 2: 0 Mbps Class 3: 12 × 10/100 = 1.2 Mbps That is, for the unprocessed class 3 in the next control section So that the packet can be processed (1.2
Mbps) is preferentially allocated.

Therefore, in this embodiment, for a class having a high real-time property, a band sufficient to satisfy the required delay time can be secured in advance. In addition, for a class that has a large amount of use (determined by the number of packets) in a recent certain time section, priority is further given to increase the throughput, and priority control for shortening the delay time is dynamically performed within a minute time that is close to real time. Can be realized.

In addition, a minimum bandwidth can be allocated to a class that uses a small number of passing packets and uses a small number of packets so that the delay time in each node (router) does not exceed a certain limit.

[0054]

According to the present invention, a band can be allocated so that a packet of a frequently used class is given priority. In addition, the bandwidth of the packet of the class can be allocated so as to satisfy the required quality of the application with high real time. Even for a class with low real-time property, it is possible to make the average throughput for a relatively long time equal to or higher than a certain value.

By employing the apparatus of the present invention, the IP
Since it becomes possible to allocate a band according to the frequency of use to a user or an application to use a network or the like, it is possible to further increase the traffic in the network and activate the use of the network.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating a packet scheduling process according to an embodiment.

FIG. 2 is a block diagram illustrating a configuration of a router that implements the present invention.

FIG. 3 is a time chart showing the timing of processing according to the embodiment;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 Packet information collection part 11 Class determination part 12 Header reading measurement part 13 Packet header analysis / calculation part 14 Routing table 15 Header write control part 16 Traffic characteristic quantity calculation part 17 Allocation band calculation part 18 Processing class indication part 19 Database 20 Scheduling part 21 queue buffer 22 packet processing unit 31 clock 32 timer

Claims (4)

[Claims] 1. A packet scheduling method for storing sequentially input packets as a queue and determining the order of packets to be taken out of the queue for processing, comprising the steps of: And classifying the plurality of classes into a priority group corresponding to stream type traffic or interactive type traffic and a non-priority group corresponding to bulk transfer type traffic. A first band sufficient to satisfy the required delay time of the class is preferentially allocated to a queue that stores packets of the class belonging to the priority group in a queue, The result of subtracting the sum of the first band from the value of is defined as a second band, For each class belonging to the priority group, the frequency of use in a specific period is detected for each class. For a queue holding packets of each class belonging to the non-priority group, the frequency of use detected in the latest specific period is used. Dynamically assigning a part or the entirety of the second band with a weight according to the difference between the queues, and determining the order of the packets extracted from the plurality of queues according to the size of the band assigned to each queue. A packet scheduling method characterized by the following. 2. The packet scheduling method according to claim 1, wherein when there is a specific class in which the allocated bandwidth is too small in the non-priority group, priority is given to a class belonging to the priority group. First, a third band is first allocated to a queue holding packets of the specific class, and then the first band is allocated to a queue holding packets of a class belonging to the priority group. A packet scheduling method characterized by assigning. 3. The packet scheduling method according to claim 1, wherein the use frequency of each class of the non-priority group is detected based on a number of packets arriving in the specific period and an average packet length of the arriving packets. Characteristic packet scheduling method. 4. A packet scheduling apparatus for accumulating sequentially input packets as a queue and determining the order of packets to be taken out of the queue for processing, comprising: Classifying means for classifying the plurality of classes into a priority group corresponding to stream-type traffic and a non-priority group corresponding to bulk transfer-type traffic; and a queue independent of each other for packets of each class. A first bandwidth sufficient to satisfy the required delay time of the corresponding class is preferentially allocated to a queue for holding packets of the class belonging to the priority group. And a result obtained by subtracting the sum of the first band from the value of the entire band. Bandwidth calculating means for determining a second bandwidth, use frequency detecting means for detecting the use frequency in a specific period for each class belonging to the non-priority group, and holding packets of each class belonging to the non-priority group. A second band allocating unit for dynamically allocating a part or the whole of the second band with a weight corresponding to a difference in use frequency detected in a latest specific period, to the queue; Scheduling means for determining the order of packets taken out of the queue according to the size of the bandwidth allocated to each queue.