JP4447521B2 - Packet scheduler and packet scheduling method - Google Patents

Description

  The present invention relates to a packet scheduler, and more particularly, to a packet scheduler and a packet scheduling method for effectively guaranteeing bandwidth during packet transfer.

  In the packet scheduler, when traffic from a plurality of users is multiplexed on a single line, if there is an input exceeding the line speed to be output, congestion occurs and packet discard due to buffer overflow occurs. At this time, in order to avoid the influence of a certain user's bandwidth on the bandwidth of other users, a guaranteed bandwidth that can be used for each user is set, and a fair bandwidth that allows the set guaranteed bandwidth to be used without fail. Control is required.

  On the other hand, QoS (Quality of Service) control such as bandwidth guarantee and delay time guarantee is required for the packet scheduler in order to comfortably use an application that requires real-time characteristics such as audio and video via the network.

  Under such circumstances, there are Patent Document 1 and Non-Patent Document 1 shown below as well-known documents disclosing a queue scheduling method for QoS control. In these documents, a WFQ (Weighted Fair Queuing) method is used in which an output completion scheduled time is calculated according to a weight proportional to the arrival packet length and the user's guaranteed bandwidth, and the output completion scheduled time is output in order from the smallest packet. A packet scheduling method based on the above is disclosed.

  For example, in the technique disclosed in Patent Document 1, queues of (number of users) × (number of priority classes) are provided, and arrived packets are distributed to each queue according to users and priority classes. While the PQ (Priority Queueing) method is used for the class queue, the WFQ method in which weighting is proportional to the total guaranteed bandwidth of all the priority classes for each user is used for each user. The bandwidth is guaranteed with the total bandwidth.

  In the technique disclosed in Non-Patent Document 1, a high-priority class queue for transferring packets with a short delay time and a (number of users) × (priority class number other than high priority) queues are provided. The packet having the highest priority among the arriving packets is stored in the high priority class queue in the order of arrival together with all the users, and is output with the highest priority by the PQ method. On the other hand, packets of other priority classes are distributed to queues according to users and priority classes, the PQ method is used for queues of multiple priority classes within the same user, and users other than the high priority class for each user. By using the WFQ method weighted in proportion to the guaranteed bandwidth in each class, the bandwidth for each user in the low priority class is guaranteed.

JP 2001-197110 A Konomi, Kensaku Amaha, Kazuyo Miyamoto, Akira Nakago “High-Speed Packet Scheduling Method Based on WFQ” IEICE Technical Report SSE98-78, IN98-59, CS98-75, September 1998, P37- 44

  However, in the packet scheduling method using the WFQ method as shown in Patent Document 1, the bandwidth for each user is guaranteed, but the read delay of the packet stored in the high priority class queue outputs the number of users and the number of users. The delay time increases in proportion to the packet length, and in the worst case, a delay time of (number of users-1) × (maximum packet length) occurs. Therefore, if a large number of users, such as thousands to tens of thousands, use an application that requires real-time performance such as audio and video, even if a high-priority class is set, the delay time is large and it cannot be used comfortably. There was a problem that there was a possibility.

  Further, in the packet scheduling method using the WFQ method as shown in Non-Patent Document 1, the high priority class packet is preferentially output by the PQ method, so that even if the number of users increases, Patent Document 1 The delay does not become large. However, since weighting in proportion to the guaranteed bandwidth other than the high priority class is performed for each user by the WFQ method, there is a problem that it is not possible to guarantee the bandwidth for each user combining the high priority class and the other classes. It was.

  The present invention has been made in view of the above, and reduces the delay time of packets of a high priority class, and combines the fairness of bandwidth guarantee and delay time guarantee, and the combination of a high priority class and a low priority class. An object of the present invention is to provide a packet scheduler and a packet scheduling method that can realize bandwidth guarantee for each user.

In order to solve the above-described problems and achieve the object, the packet scheduler according to the present invention determines the output order of packets stored in the high-priority class queue and the low-priority class queue divided for each user. A scheduler that identifies a priority class and a user of an input packet and distributes them to either the high-priority class queue or the low-priority class queue; and is output from the high-priority class queue A packet length notifying unit for notifying the user and the packet length of the packet to be transmitted, each user and packet length information notified from the packet length notifying unit, the packet length of the packet stored in the low priority class queue, and the user High priority class and low priority class assigned to each A WFQ control unit that calculates a scheduled output completion time of a packet stored in the low priority class queue at a predetermined timing for each user based on a weight value determined for each user according to the total bandwidth of Among packets stored in the low priority class queue and output via the WFQ control unit and packets stored in the high priority class queue, the packets stored in the high priority class queue are given priority. An output PQ control unit, and the WFQ control unit adds a new packet to the low-priority class queue that is empty when the next packet to be output exists in the low-priority class queue from which the first packet is output. The output completion scheduled time is calculated at any timing when a new packet arrives .

  According to the packet scheduler of the present invention, the user and the packet length of the packet output from the high priority class queue are notified, and each information of the notified user and packet length and the packet stored in the low priority class queue are notified. Completion of output of packets stored in the low-priority class queue based on the packet length information and the weight value determined for each user according to the total bandwidth of the high-priority class and low-priority class assigned to each user Since the scheduled time is calculated for each user at a predetermined timing, the delay time of the priority class packet is reduced, the fairness between the bandwidth guarantee and the delay time guarantee, and the high priority class and the low priority class. The combined band guarantee can be realized for each user.

  Embodiments of a packet scheduler and a packet scheduling method according to the present invention will be described below in detail with reference to the drawings. In addition, this invention is not limited by this embodiment.

Embodiment 1 FIG.
FIG. 1 is a block diagram illustrating a functional configuration and a queue configuration of the packet scheduler according to the first exemplary embodiment of the present invention. The packet scheduler shown in FIG. 1 includes a packet distribution control unit 10, a high priority class queue 11, a low priority class queue 12 (12 ₁ , 12 ₂ ,..., 12 _n ), a packet length notification unit 13, and a WFQ control unit 14. And a PQ control unit 15.

Next, the function of each component shown in FIG. 1 will be described. In the figure, the packet distribution control unit 10 recognizes the priority class and user (# 1 to #n) of the input packet, and sends the input packet to each queue (high priority class queue 11 or low priority class queue). To any one of 12 queues). The high priority class queue 11 stores the packets of the high priority class distributed by the packet distribution control unit 10 in the order of input. On the other hand, the low priority class queue 12 stores the low priority class packets distributed for each user by the packet distribution control unit 10. The packet length notification unit 13 recognizes the user and the packet length (number of bytes) of the packet output from the high priority class queue 11 and notifies the WFQ control unit 14 of it. In FIG. 1, n low priority class queues 12 _{1 to} 12 _n for n users of one class are illustrated, but when there are a plurality of low priority classes, (number of users) × ( Queues for the number of low priority classes) are provided.

The WFQ control unit 14 performs the following processing.
(1) The packet length (number of bytes) of the high priority class notified from the packet length notification unit 13 is added and held for each user.
(2) When a packet arrives in an empty queue or when a low-priority class packet is output, add the packet length of the packet that becomes the head of the queue and the held high-priority class packet length. To do.
(3) The added output calculated in the process of (2) is weighted in proportion to the total bandwidth of the high priority class and the low priority class for each user, and the added packet length and the user weight value ( The scheduled output completion time of each low priority class queue 12 is calculated based on the (weight value).
(4) Read the packet with the shortest scheduled output completion time from the top packets of each low priority class queue 12.

  The PQ control unit 15 gives priority to the packets stored in the high priority class queue 11 and outputs the low priority class queue 12 selected by the WFQ control unit 14 when the high priority class queue 11 is empty. Output the first packet.

  By the way, the packet scheduler of this embodiment is characterized in that it performs the operation of WFQ control in which the high priority class and the low priority class cooperate. More specifically, among the bands allocated to the high-priority class, the high-priority class is included by performing linkage processing that transfers the unused band (surplus band) to the low-priority-class band for each user. This is to guarantee a set bandwidth for each user. Therefore, in order to facilitate understanding of the present invention having such characteristics, the packet scheduler described in Non-Patent Document 1 described above will be described, and then the operation of the packet scheduler according to the present embodiment will be described. . 2 is a diagram illustrating a configuration of the packet scheduler described in Non-Patent Document 1 and the like, and FIG. 3 is a diagram for explaining a control operation of the WFQ control unit 133 illustrated in FIG.

2, the packet scheduler shown in FIG. 2 includes a packet distribution control unit 130, a high priority class queue 131, a low priority class queue 132 (132 ₁ , 132 ₂ ,..., 132 _n ), and a WFQ control unit. 133 and a PQ control unit 134.

  The packet distribution control unit 130 distributes input packets to queues corresponding to users (# 1 to #n) and priority classes. The high priority class queue 131 stores the high priority class packets of all users in the order of input. On the other hand, the low priority class queue 132 stores a low priority class packet for each user. The WFQ control unit 133 performs WFQ control according to the weight (weight) proportional to the guaranteed bandwidth of the low priority class set for each user and the head packet length (number of bytes), and outputs a packet from each queue. The PQ control unit 134 prioritizes and outputs the packets in the high priority class queue 131 over the low priority class queue 132 selected by the WFQ control unit 133.

  In FIG. 2, there are a high-priority class queue for transferring packets with a short delay time and a queue of (number of users) × (number of priority classes other than high priority (only one class in FIG. 2)). A configuration is adopted in which a packet having the highest priority among the arrived packets is output in the order of arrival in the PQ method so that the packet transmission with a short delay time is performed. On the other hand, priority class packets other than the high priority class are distributed to queues according to the user and the priority class, and a plurality of priority class queues are output by the PQ method within the same user (however, only one class is shown in FIG. 2). However, the PQ control unit is not shown in the figure because it is classified into two types). On the other hand, for each user, the data is output by the WFQ method in which weighting in proportion to the guaranteed bandwidth in the class other than the high priority class is performed for each user. As a result of such processing, the band for each user is guaranteed only in the low priority class.

  Next, the output operation of the WFQ control unit 133 shown in FIG. 2 will be described with reference to FIG. In addition, about the component same as FIG. 2, the same code | symbol is attached | subjected and shown.

In FIG. 3, the WFQ control unit 133 includes the minimum guaranteed bandwidth ratios φ1 to φn assigned to the low priority class queues 132 _{1 to} 132 _n storing the low priority class packets and the packets arriving at the respective queues. The packet output completion scheduled time is calculated on the basis of the length, and the _first packet stored in the low priority class queues 132 _{1 to} 132 _n under the restriction of the bandwidth R, in order from the packet with the smallest output completion scheduled time. Output the packet. Here, Fi (k) (F1 (1), F1 (2), F2 (1),..., Fi (1),..., Fn (1)) shown in FIG. Represents the scheduled output completion time of the k-th packet from the beginning.

  Next, a method for calculating Fi (k) will be described. For example, when a k-th packet from the head arrives at queue #i at a certain time t, the total weight sum φsum of the queues in which packets are accumulated, including the queue where the packets arrived, is given by expressed.

    φsum = Σ {φi | queue where i∈packet exists} (1)

  Further, assuming that the read bandwidth is R, the output bandwidth Ri of the queue #i is expressed by the following equation.

    Ri = R × φi / φsum (2)

  Note that φsum represented by Expression (1) is an element that varies depending on the packet arrival order k. As a result, Ri represented by Expression (2) is also a function of k, and is expressed as Ri (k). .

  Also, assuming that the length of the arrived packet is Li (k), the scheduled output completion time Fi (k) of the arrival packet is determined by the following equation.

Fi (k) = Fi (k-1) + Li (k) / Ri (k) (when k> 1)
Fi (k) = Fi (1) = t + Li (1) / Ri (1) (when k = 1) (3)

  Further, the following packets are output from the scheduled output completion time Fi (1) of the first packet of each queue.

    min {Fi (1) | queue where iεpacket exists} (4)

  When two or more packets are accumulated in the queue to which the packet is output, the output completion scheduled time of the accumulated packet is updated as shown in the following equation each time the leading packet is output.

    Fi (j + 1) → Fi (j) (1 ≦ j ≦ k−1) (5)

  If only one packet is accumulated in the output queue, the queue becomes empty after being read out. Therefore, when a new packet arrives, a new Fi (1) is obtained based on the equation (3). Is determined.

  With the above processing, the fairness of the bandwidth for each queue is maintained. Further, since the output order of the packets in each queue is output in ascending order of the output completion scheduled time, fairness is maintained. However, since weighting in proportion to the guaranteed bandwidth other than the high-priority class is performed for each user by the WFQ method, as described in the section of the above problem, for each user who combines the high-priority class and other classes. Cannot guarantee the bandwidth.

  This is because, for example, when a certain user uses a high priority class with a bandwidth less than the contract with the high priority class, the surplus bandwidth cannot be used by the user in a class other than the high priority class. This means that all users share the bandwidth ratio allocated to each user other than the high priority class. Therefore, in the configuration shown in FIG. 2, it is not possible to guarantee the bandwidth by the total bandwidth of the high priority class and the other classes for each user.

  By the way, in the above-described idea, every time a packet is output from the WFQ control unit 133, the estimated output completion time of the arrival packet is calculated based on the equations (1) to (4), and the leading packet is output. It is necessary to update the scheduled output completion time of the packets accumulated in the queue based on Expression (5), and there is a disadvantage that the processing load increases when the number of queues n is large. Therefore, a method for reducing the processing load by adopting a simple calculation method as described below is known.

In this method, first, the timing for calculating the scheduled output completion time is limited to the following two points.
(1) When there is a packet to be output next in the queue from which the first packet was output at time t (2) When a new packet arrives in the queue that was empty at time t

  In addition, in order to eliminate the division by hardware, the weights of the respective queues are determined in advance such that the total φall of the weighting factors of all the queues satisfies the following expression.

    φall = 1 = Σφi (1 ≦ i ≦ n, 0 ≦ φi ≦ 1) (6)

  Also, the weight value (reciprocal number) Wi to be set is defined as follows:

    Wi = 1 / (R × φi) (7)

  If Expression (7) is used, the scheduled output completion time Fi (1) of the first packet in the queue #i is expressed as the following expression.

    Fi (1) = t + Li (1) × Wi (8)

  In addition, a packet having the following condition is output from the scheduled output completion time Fi (1) of the first packet of each queue (reprinted in Expression (4)).

    min {Fi (1) | queue where iεpacket exists} (9)

  With such a simple calculation method, every time a packet is output or every time a packet arrives in an empty queue, the processing based on Expression (8) and Expression (9) may be performed. Compared with 1) to (5), the amount of calculation is reduced, and the processing time can be shortened.

  Note that it is free to use the normal method according to equations (1) to (5) or the simple method according to equations (8) and (9), and the number of queues and packets that arrive per unit time. It may be determined based on the number or the like.

  Next, the operation of the packet scheduler shown in FIG. 1 will be described with reference to FIG. FIG. 4 is a diagram for explaining the WFQ control operation of the low priority class in cooperation with the high priority class in the packet scheduler of the first embodiment shown in FIG.

In FIG. 4, when a packet of the high priority class of user #i arrives at the high priority class queue 11, the packet is output first by the PQ method. In this case, in order to guarantee the bandwidth with the total bandwidth of the high priority class and the low priority class for each user, the bandwidth in which the packet of the high priority class is read from the packet of the low priority class queue 12 _i of the same user #i It is necessary to lower the priority by delaying the scheduled output completion time. Therefore, when a packet arrives in a queue where the low-priority class queue is empty, or when there is a new head packet from which the top packet of the low-priority class queue is output, the high-priority class queue 11 Using the packet length Lhi (1) of the packet of user #i output from, the output completion scheduled time Fi (1) of the first packet in the low priority class queue represented by equation (8) is corrected as follows: To do.

    Fi (1) = t + {Li (1) + Lhi (1)} × Wi (10)

  In order to perform such WFQ control, a reciprocal number Wi of the weight proportional to the total bandwidth of the high priority class and the low priority class is set for each user, and the user and the packet length output from the high priority class queue are always set. Need to recognize. Further, when a plurality of packets of the same user #i are output from the high priority class queue, it is necessary to add the packet lengths to the low priority class packet length Li (k).

  1 and 4 exemplify a case where one high priority class queue corresponding to one high priority class is provided, but there are a plurality of high priority classes, and the high priority class queue is a high priority class. Each time a packet of user #i is output from each high-priority class queue, the packet length is added to the queue where user # i's low-priority class queue is empty. When a packet arrives, or when the first packet in the low priority class queue is output and there is a new leading packet, it has been output from each high priority class queue of user #i by that time The scheduled output completion time may be calculated using the total addition value of the packet length.

  In such a case, if the total addition value of the packet length output from each high priority class queue of the user #i is Lhisum (1), the equation (8) is modified as the following equation.

    Fi (1) = t + {Li (1) + Lhisum (1)} × Wi (11)

  Expression (11) means that the scheduled output completion time is delayed by the bandwidth (packet length) output from the high priority class queue for each user. By processing in this way, the high priority class and the low priority are set. The bandwidth set for each user can be guaranteed by the total bandwidth of the classes. In addition, since the high priority class packet is read by the PQ method, it can be output with a low delay regardless of the number of users.

  As described above, according to the packet scheduler and the packet scheduling method of this embodiment, a high-priority class queue that outputs with the lowest delay and the highest priority can be used for applications that require real-time performance in the network. In addition, the reciprocal number Wi of the weight proportional to the bandwidth (minimum bandwidth) of all classes in the network is set for each user, and the bandwidth output in the high priority class is considered for the low priority class queue. By performing WFQ control for each user, it is possible to guarantee a fair bandwidth for each user without squeezing the bandwidth of other users who use the same network. In addition, since bandwidth control based on the WFQ method is performed, fairness can be ensured for the output order of the low priority class packets.

Embodiment 2. FIG.
In the first embodiment, when a packet is output from the high priority class queue, the packet length output from the high priority class queue to the corresponding user is set to the empty low priority class queue, or When the first packet in the low priority class queue is output and there is a new leading packet, the bandwidth for each user is guaranteed by adding to the packet length of the low priority class packet.

  However, as in the former case, there is no packet in the low priority class queue of the corresponding user at the time when the packet is output from the high priority class queue, and a low priority packet arrives after a long time has passed. If the packet for the minimum bandwidth guarantee has not been output at a certain time, the packet length output in the high priority class is added when the first packet arrives in the low priority class queue. Is excessively lowered, and there is a problem that fairness is not satisfied.

That is, Embodiment 1
(1) When there is always a packet in the user's low-priority class queue (2) When the total bandwidth of the user's high-priority packet and low-priority packet is input above the minimum guaranteed bandwidth (3) High-priority class queue Under certain conditions, such as when no high-priority packet is input, it is possible to guarantee the bandwidth and fairness accurately for each user, but under other conditions, the bandwidth for each user may not be guaranteed.

  On the other hand, in this embodiment, even when there is no packet in the low-priority class queue of the corresponding user at the time when the packet is output from the high-priority class queue, WFQ control that enables bandwidth guarantee for each user It shows about. Note that the queue configuration and functional configuration are the same as or equivalent to those in FIG.

  Next, the operation of the packet scheduler according to this embodiment will be described with reference to FIGS. 5, 6-1 and 6-2. FIG. 5 is a diagram for explaining the WFQ control operation of the low priority class in cooperation with the high priority class in the packet scheduler of the second embodiment. FIGS. 6A and 6B are reference times for lowering the priority in the low priority class by the bandwidth of the mth packet when m packets are output from the high priority class queue. It is a figure which shows an example when each differs.

  In FIG. 5, when there are no packets in the low priority class queue of user #i, m packets of user #i are output from the high priority class queue, and the time when the mth packet is output is th (m). Suppose the packet length is Lhi (m). Now, in order to lower the priority of the low-priority class of the user #i by the bandwidth of the m-th packet output from the high-priority class queue, the packet length Lhi (m ) Is accumulated. This concept is the same as in the first embodiment. At this time, VFi (m−1) is assumed to be VFi (m−1) as the virtual output completion scheduled time in the low priority class by the packet output from the high priority class queue by the previous (m−1) th. It compares th (m), which is the mth packet output time.

  As shown in FIG. 6A, in the case of VFi (m−1) ≧ th (m), in the high priority class up to the (m−1) th, the low priority class until the time of VFi (m−1). It is necessary to virtually lower the priority, and it is necessary to lower the priority of the low priority class by an amount corresponding to the time corresponding to the packet length Lhi (m) output m-th with reference to this time. On the other hand, as shown in FIG. 6B, in the case of VFi (m−1) <th (m), it is confirmed that the packets up to the (m−1) th are input in a band less than the minimum guaranteed band. Therefore, the time corresponding to the packet length Lhi (m) that is output in the mth order is based on the time th (m) at which the mth packet is output without using the time of VFi (m−1) as a reference. It is sufficient to lower the priority of the low priority class by that amount.

  As a result, the virtual output completion scheduled time VFi (m) of the mth virtual packet is compared with VFi (m−1) and th (m), and the bandwidth of the mth packet is added to the larger time. This VFi (m) can be expressed by the following equation.

VFi (m) = max [VFi (m−1), th (m)] + Lhi (m) × Wi
(When m> 1)
VFi (m) = th (1) + Lhi (1) × Wi
(When m = 1) (12)

  Further, when there is no packet in the low priority class queue of user #i and m packets of user #i are output from the high priority class queue, at time t in the WFQ control unit of the low priority class queue, When a packet with a packet length Li (1) that newly becomes the head arrives at the low-priority class queue of user #i, the expected output completion time Fi (1) of this head packet can be expressed by the following equation.

    Fi (1) = max [VFi (m), t] + Li (1) × Wi (13)

  As described above, according to the packet scheduler and the packet scheduling method of this embodiment, a weight proportional to the total bandwidth of the high priority class and the low priority class is set for each user, and packets are output from the high priority class queue. The priority is virtually lowered by the bandwidth (packet length) output from the high priority class queue based on the user information and packet length of the high priority class, and the minimum bandwidth is guaranteed when the user is at a certain point. Since the reference time for virtually lowering the priority is varied based on the information indicating whether or not the minute packets have been output, in addition to the effects of the first embodiment, for example, in the low priority class queue A packet exists in the low priority class queue when there is no packet for a long time or when a high priority class packet is sent Be an the absence, it is possible to band warranty to the user unit.

Embodiment 3 FIG.
FIG. 7 is a diagram illustrating a functional configuration of the WFQ control unit according to the third embodiment of the present invention, and illustrates a functional configuration embodying the WFQ control unit 14 included in the packet scheduler illustrated in FIG. .

  The WFQ control unit shown in the figure includes a low priority class priority calculation unit 81, a fractional memory 82, a time register 83, an adder 84, a packet information storage scheduled output completion time FIFO (First In First Out) memory 85, an output. Each component includes a control unit 86, a high priority class priority calculation unit 87, and a high priority band memory 88.

In FIG. 7, the low priority class priority calculation unit 81
(1) When a new packet arrives in an empty low-priority class queue (2) When the first packet of a certain low-priority class queue is output and there is a next packet, the queue number of that packet under two conditions i, based on the packet length Li (1) and the weighting factor Wi of the queue, the calculation of “Li (1) × Wi” is performed, and the priority of the packet of the low priority class is output.

  The fraction memory 82 is determined by the increment of the packet information storage output completion time FIFO memory 85 in order to reduce the circuit size among the output completion scheduled times of the corresponding packets of the low priority class determined by addition by the adder 84. Holds the fraction of the scheduled output completion time to be rounded down in queue units. The fraction information stored in the fraction memory 82 is carried over to the next packet to be output. For example, in the case of (1), the fraction of the corresponding queue is 0, and in the case of (2), the fractional value of the corresponding queue stored in the fraction memory 82 is output to the adder 84, It is overwritten by a new fractional value that is generated when it is added by the adder 84.

  The high-priority class priority calculation unit 87 outputs a packet from the high-priority class queue (however, when the high-priority class is input as a PQ-type high-priority queue, the packet may arrive) Then, (Lhi × Wi) is calculated from the queue number i of the packet, the packet length Lhi, and the weighting factor Wi of the corresponding queue. The high-priority bandwidth memory 88 sequentially adds the values of (Lhi × Wi) notified from the high-priority class priority calculation unit 87 in units of low-priority class queues, and total priority of the high-priority classes (Lhisum × Wi). In the above cases (1) and (2), the value of (Lhisum × Wi) of the corresponding queue is output to the adder 84 in units of low priority class queues, and the value of the corresponding queue is output after the output of the corresponding queue. To clear. The time register 83 holds the time t when the packet is output from the packet information storage output completion scheduled time FIFO 85.

  In the above cases (1) and (2), the adder 84 uses the low-priority class priority from the low-priority class priority calculator 81, the fractional value from the fraction memory 82, and the high-priority bandwidth memory. The total priority value from 88 and the time t of the time register 83 are added to determine the scheduled output completion time, and among the corresponding queue number i, the packet length Li (1), and the added scheduled output completion time The upper bits that can be handled by the packet information storage output completion time FIFO memory 85 are output to the packet information storage output completion time FIFO memory 85, and the lower bits that are the remaining fractions are output to the fraction memory 82. The packet information storage output completion scheduled time FIFO memory 85 stores the information output from the adder 84. The output control unit 86 provides a register indicating the presence / absence of packet information for each FIFO memory, and scans the FIFO memory to search the FIFO memory having the smallest output completion scheduled time and storing the packet information. The packet output order is determined based on the queue number stored in the memory, and the output timing is determined based on the packet length.

  Here, in the operation of the high-priority bandwidth memory 88 described above, when there are no packets in the low-priority class queue of the same user and many packets are output (arrive) from the high-priority class queue, When the low priority class queue arrives, the scheduled output completion time is delayed by the bandwidth of the high priority class queue that has been output so far. May temporarily decrease. In addition, the longer the idle time of the low priority class queue, the larger the capacity of the high priority bandwidth memory 88 for each queue is required.

  Therefore, in the case of (1), the high priority band memory 88 outputs 0 to the adder 84 as in the fraction memory 82, and only in the case of (2) above, the high priority class queue. It is preferable to output a value obtained by adding these bands. When such a process is performed, in the case of (1) above, there is a possibility that the corresponding user's bandwidth may be output momentarily, but considering that the low priority class queue was empty until then. Thus, it is unlikely that congestion will occur in the low priority queue. In the case of (1), the value stored in the high priority band memory 88 can be cleared, so that the capacity of the high priority band memory 88 may be small.

  As described above, according to the WFQ control unit in the packet scheduler of this embodiment, various storage units such as a fraction memory, a high priority band memory, a time register, a high priority class priority calculation unit, and a low priority Since various arithmetic units such as a class priority arithmetic unit and an adder are provided and information stored in these components is appropriately managed, the circuit scale is reduced and the processing time is shortened. At the same time, WFQ control capable of ensuring fairness of bandwidth control for a large number of users considering high priority class bandwidth is also realized.

  In this embodiment, the packet information storage output completion scheduled time FIFO memory 85 is considered so as to have a fractional memory 82 in consideration of the bit increments. However, the bit increments of the FIFO memory are included. Is sufficient, the fractional memory 82 may be omitted.

  As described above, the present invention is useful as a packet scheduler having a WFQ control function capable of ensuring fairness of bandwidth control among a large number of users considering high priority class bandwidths.

It is a block diagram which shows the function structure and queue structure of the packet scheduler concerning Embodiment 1 of this invention. It is a figure which shows the structure of the packet scheduler shown by the nonpatent literature 1 etc. It is a figure for demonstrating control operation of the WFQ control part 133 shown in FIG. 6 is a diagram for explaining a WFQ control operation of a low priority class in cooperation with a high priority class in the packet scheduler of Embodiment 1. FIG. FIG. 10 is a diagram for explaining a WFQ control operation of a low priority class in cooperation with a high priority class in the packet scheduler of the second embodiment. When m packets are output from a high priority class queue, it is a figure which shows an example when the reference | standard time at the time of lowering the priority in a low priority class by the band of the said mth packet differs (VFi ( If m−1) ≧ th (m), th (m): the time when the mth packet was output, VFi (m−1): output from the high priority class queue before (m−1) th Virtual output completion scheduled time in low priority class by packet). When m packets are output from a high priority class queue, it is a figure which shows an example when the reference | standard time at the time of lowering the priority in a low priority class by the band of the said mth packet differs (VFi ( If m-1) <th (m), th (m): time when the mth packet was output, VFi (m-1): output from the high priority class queue before (m-1) th Virtual output completion scheduled time in low priority class by packet). It is a figure which shows the function structure of the WFQ control part concerning Embodiment 3 of this invention.

Explanation of symbols

10, 130 Packet distribution control unit ₁₁ , 131 High priority class queue 12, 12 ₁ , 12 ₂ , 12 _i , 12 _n , 132, 132 ₁ , 132 ₂ , 132 _n Low priority class queue 13 Packet length notification unit 14, 133 WFQ control unit 15,134 PQ control unit 81 Low priority class priority calculation unit 82 Fractional memory 83 Time register 84 Adder 85 Packet information storage scheduled output completion time FIFO memory 86 Output control unit 87 High priority class priority calculation unit 88 High priority bandwidth memory

Claims (5)

A packet scheduler for determining an output order of packets stored in a high priority class queue and a low priority class queue divided for each user,
A packet distribution control unit that identifies a priority class and a user of an input packet and distributes them to either the high priority class queue or the low priority class queue;
A packet length notifying unit for notifying a user and a packet length of a packet output from the high priority class queue;
Each information of the user and the packet length notified from the packet length notification unit, the packet length of the packet stored in the low priority class queue, the high priority class and the low priority class assigned to each user A WFQ control unit that calculates a scheduled output completion time of a packet stored in the low-priority class queue based on a weight value determined for each user according to a total bandwidth at a predetermined timing for each user;
Among the packets stored in the low priority class queue and output via the WFQ control unit and the packets stored in the high priority class queue, the packets stored in the high priority class queue are output with priority. A PQ control unit,
Equipped with a,
The WFQ control unit is either when a packet to be output next exists in the low priority class queue from which the first packet is output, or when a new packet arrives in the low priority class queue that was empty A packet scheduler characterized in that the output completion scheduled time is calculated at a timing . When m packets of a predetermined user are output from the high-priority class queue, the time when the m-th packet is output is defined as th (m), and is output from the high-priority class queue by (m-1) th. When the virtual output completion scheduled time calculated as a virtual accumulation in the low priority class of the predetermined user by the packet is VFi (m−1),
The WFQ control unit
A reference time for lowering the priority of the low-priority class of the predetermined user by the bandwidth of the m-th packet output from the high-priority class queue is set to the VFi (m−1) and the th (m). The packet scheduler according to claim 1, wherein the packet scheduler is determined based on a magnitude relationship. The WFQ control unit
A calculation means for calculating a priority applied to the output packet when the packet stored in the high priority class queue is output from the high priority class queue;
Storage means for adding and holding the priority of packets of a high priority class for each user;
Means for calculating the priority of the first packet of each queue of the low priority class;
A register holding the current time, and
An adder for adding the priority value of the packets of the high priority class, the priority of the first packet of the low priority class, and the current time output from the register;
A FIFO memory for storing queue information of packets stored in the low priority class queue corresponding to the addition value of the adder;
And determines the packet output order based on the queue information stored in the FIFO memory, according to claim 1, characterized in that to determine the output timing based on the packet length stored in the FIFO memory Packet scheduler. The WFQ control unit
When a packet stored in the high priority class queue is output from the high priority class queue and no packet exists in the low priority class queue of the predetermined user, the priority of the high priority class of the predetermined user is not added. The packet scheduler according to claim 3 . A packet scheduling method for determining an output order of packets stored in each of a high priority class queue and a low priority class queue divided for each user,
A packet distribution step for identifying a priority class and a user of an input packet and distributing the packet to either the high priority class queue or the low priority class queue;
A packet length notifying step for notifying the user and packet length of a packet output from the high priority class queue;
Each information of the user and packet length notified from the packet length notification unit, packet length information of the packet stored in the low priority class queue, the high priority class and the low priority class assigned to each user A scheduled output completion time calculating step for calculating a scheduled output completion time of a packet stored in the low priority class queue based on a weight value determined for each user according to the total bandwidth of
An output step for preferentially outputting the packet stored in the high priority class queue among the packets stored in the low priority class queue and the packets stored in the high priority class queue;
Only including,
The predetermined timing for calculating the scheduled output completion time is when there is a packet to be output next in the low priority class queue from which the first packet is output, or when a new packet is in the low priority class queue that was empty A packet scheduling method, wherein the packet scheduling method is a case of arrival .

Images