JP3595134B2 - Packet buffer device and packet discard control method - Google Patents

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a packet buffer device that is a network component of a packet communication network and a packet discard control method thereof.
[0002]
[Prior art]
The Internet is known as a basic information network that can communicate between many networks and terminals. The Internet is a packet communication network for communicating packets called IP packets. On the other hand, the ATM communication network is expected as a method for realizing a digital broadband public network capable of guaranteeing communication quality, but is also a kind of packet communication network for communicating fixed-length packets called cells. . As described above, packet communication is widely recognized as a promising communication technology, and research and development are being vigorously pursued by various organizations.
[0003]
In packet communication, the source terminal can change the transmission rate of a packet to change the usage rate of a line in a network. Therefore, there is an advantage that a plurality of terminals can effectively share the bandwidth resource of one line. In other words, when a line is not used by another terminal, one terminal can occupy the line speed, and when many terminals use the line at the same time, the line speed is divided among many terminals and shared. can do.
[0004]
In order to utilize such advantages positively, it is important that the packet transmission source terminal adaptively changes the packet transmission speed in accordance with the degree of line congestion. One of the protocols for realizing this is, for example, a protocol called TCP. TCP operates in a terminal that transmits and receives IP packets, indirectly knows the degree of congestion of the line by discarding the IP packet, and changes the transmission speed of the IP packet.
[0005]
Line congestion or congestion mainly occurs in a packet buffer device in a packet network. The packet buffer device is typically a router device of the Internet or an ATM switch device of an ATM communication network. Congestion in the packet buffer device occurs when packets input from a plurality of input ports go to one output port at the same time or when the output speed of the output port is smaller than the total speed of the input ports. Congestion also occurs, for example, when the output port is connected to Ethernet and the Ethernet is under heavy load. Furthermore, congestion also occurs when the arrival speed of a packet is higher than in a router that determines the route of each packet. Packets that cannot be output to the output port due to the congestion or congestion are temporarily stored in a buffer in the packet buffer device. However, if the amount of stored packets increases, the packets must be discarded.
[0006]
If a terminal implements a protocol such as TCP, upon detecting that the packet has been discarded, the terminal on the transmitting side lowers the transmission speed of the packet. As a result, the congestion is eliminated, and the buffer of the packet buffer device can have room again. When the terminal finds that the packet is transferred without being discarded, it increases the transfer speed again and tries to use the line effectively. As described above, generally, the terminal repeatedly increases and decreases the transfer rate.
[0007]
As the speed at which a terminal transmits a packet increases or the distance between a transmitting terminal and a receiving terminal increases, the packet buffer device arrives during a period from when a packet is discarded to when the packet input speed of the packet buffer device decreases. The amount of packets increases.
[0008]
A conventional packet buffer device compares the amount of accumulated packets with a threshold value and determines whether to discard the arriving packet. In this case, the amount of packets stored in the packet buffer device stays near the threshold from when the first packet is discarded until the packet input speed decreases.
[0009]
In this conventional packet buffer device, since the amount of packets arriving during the period in which the accumulated packet amount stays near the threshold value is large, all the flows arriving during this period experience packet discard almost at the same time. As a result, all the transmitting terminals simultaneously reduce the transmission speed. This phenomenon is known as a global synchronization phenomenon, and has caused a reduction in the utilization rate of the output port.
[0010]
As a method for solving this problem, Sally Floyd and Van Jacobson have described "Random Early Detection Gateways for Congestion Aidance", IEEE / ACM Transactions news. 1, no. 4, August 1993. There is a RED that is proposed in. RED is a method of randomly arriving packets with a probability corresponding to the average accumulated packet amount. Since packets are discarded at random, there is an effect of preventing global synchronization.
[0011]
However, in the case of RED, if there is a terminal that does not lower the transmission speed even if the discard is detected, there is a disadvantage that the throughput of another terminal is unfairly reduced due to the influence of the terminal. The RED did not have the fault tolerance of continuing to provide appropriate services to other terminals without being affected by some terminals.
[0012]
Also, Juha Heinanen and Kalevi Kikki have proposed a method called FairBuffer Allocation (FBA). FBA is a method of fairly allocating a buffer area for each flow. Since the algorithm discards packets of a flow that occupies more buffers than other flows, FBA also has the effect of suppressing the occurrence of global synchronization. It also has fault tolerance.
[0013]
However, in the FBA, it is necessary to perform a complicated operation such as dividing the total buffer capacity by the number of connections in order to calculate the maximum available buffer amount allocated for each connection, and there is a problem that the implementation becomes complicated. In particular, it has been difficult to realize this in a packet buffer device that requires high-speed operation.
[0014]
As described above, the conventional packet buffer device has a problem that the congestion state and the non-congestion state are periodically repeated, and as a result, the throughput of the line is reduced. There are disadvantages that the throughput of other terminals is unfairly reduced due to the influence of some terminals and that the implementation is complicated.
[0015]
[Problems to be solved by the invention]
As described above, the conventional packet buffer device has a problem that the congestion state and the non-congestion state are periodically repeated, and as a result, the throughput of the line is reduced. There are disadvantages that the throughput of other terminals is unfairly reduced due to the influence of some terminals and that the implementation is complicated.
[0016]
SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and provides a packet buffer device and a packet discarding device that can suppress global synchronization that lowers the average utilization rate of output ports and that can achieve fault tolerance with a simple implementation. It is an object to provide a control method.
[0017]
[Means for Solving the Problems]
The present invention is a packet buffer device for temporarily storing and processing packets classified by group, comprising: a packet storage unit for temporarily storing arriving packets; and If the total packet storage amount obtained by adding up the packet storage amounts of the plurality of groups in the packet storage unit exceeds the first threshold, the group to which the arriving packet belongs should be discarded. Registration means for registering as a group, and when a packet storage amount in the packet storage means for each of the registered groups falls below a second threshold value different from the first threshold value. Deleting means for deleting the registration made for the group, and discarding packets belonging to the registered group. Characterized in that a disposal unit.
Preferably, the discarding unit may discard packets arriving from the group belonging to the registered group.
Preferably, the group may be classified based on header information of the packet.
Preferably, this may be done.
Also, the present invention is a packet discard control method in a packet buffer device for temporarily storing packets classified by group in a packet storage unit and processing the packets. Registering the group to which the arriving packet belongs as a group to be discarded when the total packet storage amount obtained by adding the packet storage amounts of the plurality of groups exceeds a first threshold value; And when the packet storage amount in the packet storage unit for each of the registered groups falls below a second threshold value different from the first threshold value, Deleting the registration that has been made, and discarding packets belonging to the registered group. Characterized in that it has and.
[0024]
According to the present invention, packets of a group (registered) selected as a “group to which packets are to be discarded” are selectively discarded, so that transmission terminals of other groups do not need to reduce the transmission speed, and are global. Synchronization hardly occurs. In addition, fault tolerance can be realized by evenly selecting (registering) and discarding packets in a group of flows or the like in which the transmitting terminal does not appropriately respond to the congestion state. Further, since this control can be performed by a simple process such as threshold value comparison, it is easy to realize.
[0025]
Note that the invention relating to the above-described apparatus is also established as an invention relating to a method, and the invention relating to a method is also established as an invention relating to an apparatus.
Further, the above-described invention is also realized as a machine-readable medium storing a program for causing a computer to execute a corresponding procedure or means.
[0026]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
First, basic functions provided in the packet buffer device according to the present embodiment will be described.
[0027]
In the present embodiment, control relating to packet discarding for a packet buffer to be transmitted after arriving packets are temporarily stored. In the present embodiment, the following functions are provided for that purpose.
[0028]
(1) First, in the present embodiment, the packet buffer device is provided with a function of registering and managing one or a plurality of groups selected according to a predetermined criterion as "groups to discard packets".
[0029]
The "group" here is, for example, a virtual connection in the case of ATM communication, and can be known by referring to an area called VPI or VCI of the header. In the case of IP communication, it is, for example, a flow, and is identified by, for example, a set of a destination address, a protocol identifier, a port number, and a source address, a protocol identifier, and a port number. The groups and the virtual connections or flows do not need to correspond one-to-one. For example, the destination address may be compressed into 6 bits by a method called hash and may be classified into 64 groups.
[0030]
The registration management of "groups to discard packets" may be managed as a table provided with a flag indicating whether each group is a "group to discard packets" or "to discard packets". It is also possible to create and manage a list in which identification information of a group to be a “group” is registered.
[0031]
(2) Secondly, in the present embodiment, the packet buffer device monitors whether or not congestion (congestion exceeding a standard) has occurred in a plurality of groups as a whole, and when it is determined that congestion has occurred, the packet buffer device arrives. A function is provided for additionally registering one group selected from the group of the packets and the group to which the packets stored in the packet buffer belong to the “group to discard the packet”.
[0032]
As a simple method, it can be determined whether or not congestion has occurred by checking whether or not the amount of packets stored in a plurality of groups exceeds the “threshold value”. Whether to manage the number of packets or the total length of the packets as the amount of accumulated packets may be appropriately selected depending on the resource management policy of the network. it can.
[0033]
As the accumulated packet amount, the accumulated packet amount at each point in time may be used. However, a so-called moving exponential average (a method of calculating an average used when calculating a round trip time of TCP) or the like is used. Alternatively, an averaged amount may be used. By taking the average, it is possible to suppress a malfunction due to the arrival of a packet having a burst property.
[0034]
Various methods such as a method of measuring an input packet rate can be used to determine whether or not congestion has occurred, in addition to the method based on the amount of stored packets.
[0035]
Next, when congestion occurs, various methods can be used as to which group is to be selected as a “group to discard packets”. In principle, a method of selecting such that no unfairness occurs between groups is adopted.
[0036]
As a simple method, a group of packets arriving at that time may be selected as a “group to discard packets”. According to this method, the probability of each group being selected depends on the arrival rate, so that a group having a higher arrival rate is more likely to be selected, and fairness in communication rate allocation can be expected.
[0037]
As another method, for example, a method of selecting a packet having a large amount of stored packets for each group in a packet buffer, a method of selecting a packet having a large arrival rate for each group, a method of selecting randomly, or a round robin selection A method or the like may be used.
[0038]
In addition, when selecting a group to be added to the “group to discard packets”, if one does not select a packet having a small amount of packets stored in each group in the packet buffer or a small arrival rate for each group, It is possible to avoid performing congestion notification according to a specific flow.
[0039]
(3) Thirdly, in the present embodiment, the packet of the “group to discard packets” is discarded in the packet buffer device based on a predetermined discard algorithm, and then the group is discarded. A function to delete from the "should be group" is provided.
[0040]
As a simple method of the discard algorithm, check the group to which the arriving packet belongs, and if it is registered as a "group to discard the packet", discard the packet without putting it in the packet buffer. You may make it.
[0041]
As another method, the group to which the arriving packet belongs is examined, and if it is registered as a “group to discard the packet”, the arriving packet itself is stored in the packet buffer. Instead, a method may be used in which one of the packets belonging to the same group stored in the packet buffer is discarded at random or from the oldest one. In this case, instead of discarding one packet, all packets belonging to the same group stored in the packet buffer may be discarded.
[0042]
Further, once the congestion occurs, all the packets belonging to the “packet to be discarded” stored in the packet buffer may be discarded.
[0043]
Next, conditions for deleting a group from the “packet to be discarded” include, for example, that the amount of packets in the packet buffer of the group has become smaller than the “threshold value” and that the group has been observed. It may be determined that the arrival rate has become smaller than the "threshold value". Also, if a certain amount of packets (for example, packets corresponding to the amount of stored packets at the time of registration) are output from the packet buffer after the group is registered as a “packet to be discarded,” A method of deleting from the “group to be discarded” may be used.
[0044]
Hereinafter, a more detailed embodiment of the packet buffer device having the above basic functions will be described.
FIG. 1 shows a configuration example of a packet buffer device according to the present embodiment.
[0045]
This packet buffer device includes a packet buffer 2, a group identification unit 4, a discard determination unit 6, and a management table 8.
The packet buffer 2 temporarily stores the arriving packet and then sends it out.
[0046]
When a packet arrives at the packet buffer device, the group identification unit 4 obtains a group to which the packet belongs by referring to the header of the packet. The groups are as described above.
[0047]
The management table 8 includes, for each group, a flag Flagi indicating whether or not the group i is a “group to discard packets”, and a stored packet amount (number of packets or data amount) of packets belonging to the group. Is stored. Note that i in Flgi and Cnti is a variable indicating a group.
[0048]
Here, the management table is used, but a list may be created and managed as described above.
The discard judging unit 6 monitors the input / output of the packet to / from the packet buffer 2 to set / cancel the “group where the packet should be discarded”. Should give a "group" instructions.
[0049]
In the discard determination unit 6, a counter value CntA (61 in the figure) indicating the sum of the accumulated packet amounts of all groups, and a first threshold value ThrA (62 in the figure) to be compared with the counter value CntA And a second threshold value Thr (63 in the figure) to be compared with the accumulated packet amount (counter value Cti) for each group.
[0050]
The first threshold value ThrA is used to mark the group flag Flgi, and the second threshold value Thr is used to release the mark of the group flag Flgi. Preferably, the second threshold value Thr is set to a value close to zero.
[0051]
When the packet belonging to the group i is input to the packet buffer 2, the accumulated packet amount Cti of the group i is incremented by the number of packets “1” or the data length (for example, bit length) of the packet. Can be obtained by decrementing the number of packets “1” or the data amount of the packets when the packets belonging to.
[0052]
The total sum CntA of the accumulated packet amounts of all groups is obtained by incrementing the number of packets “1” or the data length of the packet when the packet is input to the packet buffer 2, and calculating the packet number when the packet is output from the packet buffer 2. It can be obtained by decrementing "1" or the data amount of the packet. CntA can also be obtained by taking the sum of the accumulated packet amounts Cnti of group i.
[0053]
The packet buffer 2 may be used for storing packets for each group, such as WFQ (Weighted Fair Queueing), or for storing packets not for each group, such as a FIFO (First Input First Output). The invention is applicable.
[0054]
Next, the operation of the packet buffer device according to the present embodiment will be described.
Here, a case where the packet amount is counted by the number of packets will be described.
FIG. 2 shows an example of the operation of the packet buffer device at the time of inputting a packet.
[0055]
When the packet of the group i arrives, first, the counter value Cnti indicating the current accumulated packet amount of the group i is temporarily stored in the oldCnt, and the counter CntA indicating the total sum of the accumulated packet amounts of all the groups is temporarily indicated by the oldCntA. (Step S1). Subsequently, the counter value Cnti and the counter value CntA are each incremented by one (step S2).
[0056]
Next, the counter value CntA is compared with the first threshold value ThrA, and if CntA> ThrA (step S3), the flag Flgi of the group i is marked (step S4). Marking the flag Flgi means setting the flag Flgi to a state indicating that it is a "group to discard packets". Here, the content of the flag Flgi is set to 1.
[0057]
On the other hand, if CntA> ThrA is not satisfied in step S3, step S4 is not executed, and the process proceeds to step S5.
Subsequently, if the flag Flgi is marked (step S5), the arriving packet of the group is discarded (step S7), and the incremented counter value Cnti and counter value CntA in step S2 are respectively returned to the original values. Return to oldCnt and oldCntA (step S8).
[0058]
On the other hand, if the flag Flgi is not marked in step S6, the arriving packet is stored in the packet buffer 2 (step S6).
Next, the stored packet amount oldCnt of the group i before the increment is compared with the second threshold value Thr. If oldCnt <Thr or oldCnt = Thr (step S9), the flag Flgi of the group i is cleared. The mark is released (step S10). To remove the mark of the flag Flgi means to set the flag Flgi to a state indicating that it is not a “group where packets should be discarded”. Here, the content of the flag Flgi is set to 0.
[0059]
Next, FIG. 3 shows an example of the operation of the packet buffer device when outputting a packet.
When outputting the packet of the group i, first, the values of the stored packet amount counters Cnti and CntA are decremented by one (step S11).
[0060]
Next, the accumulated packet amount Cti of the group i after decrement is compared with the second threshold value Thr, and if Cnti <Thr or Cnti = Thr (step S12), the flag Flgi of the group i is cleared. The mark is released (step S13).
[0061]
The above procedure can be implemented with various changes.
For example, in the above description, the packet amount is counted by the number of packets. However, when the packet amount is counted by the data amount, in step S2 of FIG. 2, the packet is incremented by the length of the arriving packet (data length). In step S11 of 3, it is sufficient to decrement by the length of the arriving packet (data length).
[0062]
As described above, the moving exponent average may be used as the accumulated packet amount. In this case, a moving index average item for each group is added to the management table 8. In the procedure of FIG. 2, processing for saving and updating the moving index average is inserted after step S2, and the moving index average is added after step S8. A process for restoring the average is inserted, and the moving index average is a comparison target in the judgments in step S3 and step S9. In the procedure of FIG. 3, a process for updating the moving index average is inserted after step S11. In the determination of S12, the moving index average is the comparison target.
[0063]
In the above procedure, the group of packets arriving at that time is selected as the "group to discard the packet", but other methods as described above can be used.
[0064]
In addition, a method of determining whether congestion has occurred, a method of discarding a packet, a method of deleting a group from a “group to which a packet should be discarded”, and the like can also each employ the other methods described above.
[0065]
As described above, the packet buffer device according to the present embodiment discards the arriving packets of the group marked as “the group to discard packets” to make room in the packet buffer. Is less likely to be discarded. Since which group is marked fairly selected, as a result, any group is marked fairly and lowers throughput, while other groups can keep high throughput. Since all groups are unlikely to be marked at the same time, there is an advantage that the problem of global synchronization does not easily occur, and the overall throughput is higher than that of the conventional packet buffer device.
[0066]
Further, according to the present embodiment, even if there is a group that does not lower the rate despite congestion, if marked, the group is forcibly discarded to make room in the packet buffer. Can be maintained and fault tolerant.
[0067]
In the present embodiment, there is an advantage that mounting is easy because complicated numerical calculation is not required and can be realized simply by comparison with a threshold value.
According to the present embodiment, it is possible to give a packet a discard priority. For example, when there are two types of packets, a high-priority packet and a low-priority packet, an algorithm targeting only high-priority packets (eg, FIGS. 2 and 3) and an algorithm targeting only low-priority packets (eg, FIG. 2 and FIG. 3) at the same time, so that the condition for marking the flag of the low-priority packet may be earlier than the condition for marking the flag of the high-priority packet. For example, as a condition for marking a flag of a high-priority packet, a flag of a high-priority packet is marked when a total accumulated packet amount of a high-priority packet exceeds a threshold value, whereas a flag of a low-priority packet is marked. As a condition for marking the flag of the packet, if the accumulated packet amount of the high-priority packet and the low-priority packet exceeds the threshold value, the flag of the low-priority packet is marked, so that the algorithm for the low-priority packet can be set. Even if the algorithm for the high priority packet and the algorithm for the high priority packet have the same threshold value, the flag of the low priority packet is marked first.
[0068]
Note that each of the above functions can be implemented as software. Further, the present invention can be embodied as a machine-readable medium storing a program for causing a computer to execute the above-described procedures or means.
The present invention is not limited to the above-described embodiment, and can be implemented with various modifications within the technical scope thereof.
[0069]
【The invention's effect】
According to the present invention, packets of a group selected as a group to which packets are to be discarded are selectively discarded, so that transmission terminals of other groups do not need to lower the transmission speed, and global synchronization is less likely to occur. In addition, fault tolerance can be realized by evenly selecting and discarding packets even in a group in which the transmitting terminal or the like does not appropriately respond to the congestion state. Further, since this control can be performed by a simple process such as threshold value comparison, it is easy to realize.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration example of a packet buffer device according to an embodiment of the present invention.
FIG. 2 is an exemplary flowchart illustrating an example of an operation of the packet buffer device according to the embodiment when a packet is input.
FIG. 3 is an exemplary flowchart illustrating an example of an operation of the packet buffer device according to the embodiment when outputting a packet.
[Explanation of symbols]
2 ... Packet buffer
4: Group identification unit
6 ... Discard judgment part
8 Management table

Claims (4)

A packet buffer device for temporarily storing and processing packets classified by group,
Packet storage means for temporarily storing the arrived packet,
When a packet arrives, the total packet accumulated amount which is the sum of each of the packet accumulated amount of the plurality of groups in the packet storage means, when the first threshold is exceeded was the arrived packet the group including a registration means for registering as a group to be a packet discarded,
When the amount of accumulated packets in said packet accumulating means for each group to which the registration is made, below the second threshold value different from the first threshold value, performs with respect to the group Deleting means for deleting said registration ,
Packet buffer device which is characterized in that a discarding means for discarding packets belonging to the group the registration has been made. 3. The packet buffer device according to claim 1, wherein the discarding unit discards a packet belonging to the registered group and arrives. It said group includes a packet buffer device according to claim 1 or 2, characterized in that to be classified based on the header information of the packet. A packet discard control method in a packet buffer device for temporarily storing packets classified by group in a packet storage unit and processing the packets,
When a packet arrives, the total packet accumulated amount which is the sum of each of the packet accumulated amount of the plurality of groups before Symbol packet storing the unit, if it exceeds the first threshold value, and the arrival packet the group to which belongs, and the step of registering as a group should be packet discard,
When the amount of accumulated packets in said packet accumulating means for each group to which the registration is made, below the second threshold value different from the first threshold value, performs with respect to the group and the step of deleting the registration was,
Packet discard control method characterized by comprising the step of discarding a packet belonging to the group the registration has been made.

Images