JP4537937B2 - Congestion control method, congestion control program, and congestion control system - Google Patents

Description

  The present invention relates to a congestion control method, a congestion control program, and a congestion control system.

  In recent years, network flows have increased rapidly, and this trend is still ongoing. A flow is a set of packets used for the same purpose. For example, the flow is uniquely identified by the IP (Internet Protocol) address of the start point or end point of the flow, the port number of the start point or end point, the protocol, etc. Information for identification is a flow ID). Looking at the breakdown of the flow, the web and mail flows have previously occupied the majority, but in recent years there has been a report that P2P (Peer to Peer) flow has grown significantly and P2P accounts for more than half of the flow. The size of the flow is the sum of the sizes of the packets of the flow.

  The spike flow is a flow that is likely to cause congestion in the network. Spikes, for example, have a very large throughput per unit time (in other words, are bursty), and occupy a queue of communication devices in a short time, thereby causing congestion in the network. Spikes have a short survival time, but once they occur, they occur continuously, so long-term congestion tends to occur. The flow size in the network often follows a Pareto distribution, and the P2P (Peer to Peer) flow tends to follow this Pareto distribution compared to the web flow. And this feature of P2P tends to produce spikes.

  Note that spikes cannot be permanently reduced by increasing line capacity through capital investment. If spikes increase, long-term congestion is likely to occur even in a core network having a high-speed and large-capacity line. Also, the increase in line speed is higher in the access network than the core network in the low cost and easy upgrade. Therefore, even if the provider tries to avoid congestion by increasing the line capacity of the core network, congestion due to spikes continues to occur as the line capacity of the access network on the user side increases.

  And it is more efficient to control with the aim of spikes than to control the entire flow uniformly. The flow control is, for example, to lower the flow line usage rate by deleting a part of the flow packet during transfer. Also, the number of spikes is small compared to the total number of flows. In recent years, one of the features of the flow is that the flow size is Pareto-distributed, that is, a small number of flows composed of an extremely large number of packets occupy most of the total flow rate. For example, it has been observed that a flow in which the ratio of the number of flows is 0.02% of the whole occupies a band of about 60%.

  In the following, the three functions that will be referred to when the network administrator controls spikes will be described in order.

  The first function is a function of measuring the flow rate and size. Typical conventional techniques include SNMP (Simple Network Management Protocol) and NetFlow from Cisco (see, for example, Non-Patent Documents 1 and 2).

  The second function is a function for specifying a flow. The prior art has analyzed the measurement data offline to identify the flow from time to time. The flow to be identified can be determined by the operator according to the network policy. The specific target flow is defined as, for example, a flow composed of 0.1% or more of the total number of packets, or a flow whose flow size is equal to or greater than the sum of the average size of all flows and three times the standard deviation.

The third function is a flow control function. The prior art includes, for example, AQM (Active Queue Management) that preferentially discards a specific flow. As an example of the flow control function, there is WRED (Weighted Random Early Detection) that has a plurality of discard probabilities according to priority classes and randomly discards packets of low priority classes during congestion. As another example of the flow control function, there is scalability even in a large-scale network by dividing the function between a core node and an edge node, and a CSFQ (Core that calculates a fair-share rate and discards packets fairly between flows. -Stateless Fair Queuing) (for example, see Non-Patent Documents 3 and 4).
The Internet Engineering Task Force (IETF), "RFC 1157 Simple Network Management Protocol (SNMP)", [online], [October 12, 2005 search], Internet <URL: http://www.ietf.org/rfc /rfc1157.txt> Cisco Systems, “Cisco IOS software Netflow”, [online], [October 12, 2005 search], Internet <URL: http://www.cisco.com/en/US/products/ps6601/prod_presentation_list.html > D. Clark and W. Fang, “Explicit Allocation of Best-Effort Packet Delivery Service”, IEEE / ACM Trans.Net.Vol.6, August 1998, pp.362-373. Stoica, I .; Shenker, S .; Hui Zhang, “Core-stateless fair queuing; a scalable architecture to approximate fair bandwidth allocations in high-speed networks”, Networking, IEEE / ACM Transactions on Volume 11, Issue 1, Feb. 2003 Page (s): 33-46.

  However, in the conventional technology, the function for specifying the active flow for realizing the control of the active flow in real time is insufficient. An active flow is a flow during communication and is a flow that is a target of processing such as transfer and control in a communication device between a source and a destination. The control target is a flow that has a large size and is likely to be spiked among the active flows.

  In the conventional function for specifying a flow, an operator needs to specify an active flow to be controlled one by one by trial and error. However, the communication state including the active flow may change greatly with time. Therefore, if the operator's human judgment is interposed in the flow specification, the response is delayed and the active flow cannot be sufficiently controlled.

  Therefore, the users who use the network system in operation have been troubled because the use of the line is remarkably restricted by some malicious users. Annoying is caused by, for example, occupying a band or congestion caused by a spike. In addition, although the conventional AQM can perform priority control for high-rate flows, the flow size for high-rate flows is not always large, so excessive control is performed even for flows that should not be controlled. End up.

  In the conventional method, after the lifetime from the start to the end of the flow is completed, information on all packets of the flow is collected and analyzed. In the conventional method, for example, a computer different from the network system refers to log information collected by the network system offline to determine whether or not the flow is spiked. However, with this method, no countermeasures can be taken during the period when the spikes are annoying.

  In view of the above, the main object of the present invention is to solve the above-described problems and to select and efficiently control active flows that cause trouble for the network.

In order to solve the above problem, the present invention is a congestion control method in which a communication device that relays a flow discards a packet of a flow to control congestion, and is located upstream of the flow in the communication device. For each active flow passing through the communication device, the first communication device collects flow information including the accumulated flow size and stores it in the storage means, and describes the flow size in the packet header of the flow. The second communication device positioned downstream of the flow from the first communication device, and the second communication device of the communication devices is notified for each flow that passes through the device. The flow size is read from the packet header and the flow size for a predetermined flow passing through the own device in the flow size of all flows passing through the own device. By calculating a predetermined function for obtaining the drop probability of a packet from the ratio, and discard probability calculation procedure for calculating a drop probability of a packet of the given flow, according discard probability for the packet, the active flow the a given flow And a packet discarding procedure for discarding the packet. Furthermore, the present invention is a congestion control system for executing the above-described congestion control method, and the congestion control system includes a first communication device located upstream of a flow of the communication devices, and the first communication device. It is comprised by connecting with the 2nd communication apparatus of the said communication apparatus located in the downstream of a flow with a network.

As a result, the packets are discarded according to the packet discard probability calculated from the flow information so that the queue in the communication apparatus does not overflow, so that packets can be discarded fairly between flows and congestion can be avoided.
Furthermore, the flow sharing is performed by the first communication device on the upstream side of the flow with a low line speed, and the calculation of the discard probability and the packet is discarded on the second communication device on the downstream side of the flow with a high line speed. By doing so, even in a high-speed and large-capacity network that requires high-speed processing in the communication device, processing is not lost.

  The present invention is characterized in that the information collection procedure calculates the flow size by calculating a total size of packets arriving at the communication device.

  As a result, packets belonging to flows with a large flow size can be selected and discarded, so that it is possible to discard packets fairly and at the same time preferentially capture high-rate controlled flows that tend to be bursty flows. By discarding, bursty congestion can be suppressed.

  The present invention is characterized in that the information collection procedure calculates the flow size by multiplying a total number of packets arriving at the communication device by a predetermined packet size.

  As a result, it is possible to select and discard packets belonging to flows with a large number of constituent packets, so it is possible to discard packets fairly and at the same time preferentially capture high-rate controlled flow that tends to be bursty flows. It is possible to suppress burst-like congestion by discarding it to the default.

  The present invention is characterized in that the information collecting procedure collects information from all packets arriving at the communication device for a predetermined flow.

  Thereby, it is possible to perform accurate and precise congestion control by collecting information of all packets.

  The present invention is characterized in that the information collection procedure collects information from a part of all packets arriving at the communication device for a predetermined flow.

  Thereby, by sampling and processing some packets, congestion control can be performed even in a network having a high-speed and large-capacity line that requires high-speed processing because the processing catches up with the transfer rate.

The present invention, the discard probability calculation procedures, as the predetermined function, according to a continuous increasing function of the ratio was defined as the domain of the flow size, and obtains the discard probability of the packet.

  Thus, fine congestion control can be performed by changing the packet discard probability according to the continuous increase function.

The present invention is characterized in that the discard probability calculation procedure obtains the discard probability of the packet according to a step function having the flow size ratio as a domain as the predetermined function .

  Thereby, the packet discard control method can be simplified by changing the packet discard probability according to the step function.

  The present invention is a congestion control program for causing a computer to execute the congestion control method.

  Thus, the packets are discarded according to the packet discard probability calculated from the flow information so that the queue in the communication apparatus does not overflow, so that packets can be discarded fairly between flows and congestion can be avoided.

  In the present invention, the packet discard probability of the flow is calculated based on the flow information including the size of the flow passing through the communication device. As a result, a flow with a high packet discard probability is expected to be a spike, so that active flows can be selected and efficiently controlled.

  Hereinafter, a first embodiment of a network system to which the present invention is applied will be described with reference to the drawings. First, the network structure according to the present embodiment will be described.

  FIG. 1 shows a circuit-switched network such as TDM (Time Division Multiplexing) or WDM (Wavelength Division Multiplexing), or MPLS (Multi-Multicast) that labels and forwards IP (Internet Protocol), Ethernet (registered trademark), or IP packets. It is a conceptual diagram of a network of (Protocol Label Switching).

  The network is configured by connecting communication devices 9. Each communication device 9 is connected by a logical path provided by setting a lower layer path positioned relatively lower than the network layer in which the node exists. The communication device 9 may be, for example, an MPLS router of an MPLS network, a router of a packet network, or an optical cross connect of an optical network. Note that the network configuration according to the present embodiment is not limited to the circuit switching network described above, and can also be applied to a hierarchical network in which a packet network is accommodated in the circuit switching network.

  The network shown in FIG. 1 includes a communication device 9, a congestion control device 1, a transmission terminal 21, and a reception terminal 22. The flow from the transmission terminal 21 to the reception terminal 22 is configured by relaying the communication device 9 that becomes the flow path. Then, the congestion control device 1 included in the communication device 9 collects information on the flows that flow in the network, and uses the information on the flows to calculate a packet discard probability that does not cause congestion in the network. The packet is discarded according to the probability.

  Note that there may be a plurality of congestion control apparatuses 1 in one network. As a result, the number of points where congestion control is performed increases, so that congestion can hardly occur in the entire network.

  FIG. 2 is a configuration diagram of the congestion control device 1. The congestion control device 1 is configured by a combination of an information collection unit 11, a discard probability calculation unit 12, a packet discard unit 13, and a control interface 14 with an external device. The information collecting unit 11 collects information on the flows that flow in the network (see FIG. 3, details will be described later). The discard probability calculation unit 12 uses the flow information collected by the information collection unit 11 to calculate the packet discard probability so that congestion does not occur in the network. The packet discard unit 13 discards a packet of a flow to be controlled according to the packet discard probability calculated by the discard probability calculation unit 12. The control interface 14 serves as an entrance / exit for transmitting / receiving information to / from an external device.

  Further, the congestion control device 1 is configured as a computer including at least a memory as a storage unit 19 used when performing arithmetic processing and an arithmetic processing device that performs the arithmetic processing. The memory is constituted by a RAM (Random Access Memory) or the like. Arithmetic processing is realized by an arithmetic processing unit configured by a CPU 18 (Central Processing Unit) executing a program on a memory. The storage means 19 stores information (table in FIG. 3) for calculating the packet discard probability.

  FIG. 4 is an example of a flowchart for explaining the operation of congestion control according to the embodiment of the present invention. The flowchart of FIG. 4 will be described below with reference to FIGS.

  The information collection unit 11 collects flow IDs of flows that pass through the communication device 9 to which the information collection unit 11 belongs, and collects information for calculating the packet discard probability for each flow ID (S11). 3A is a table that stores information for each flow ID, and FIG. 3B is a dedicated table that stores information that does not depend on the flow ID.

  FIG. 3A shows the flow rate, the flow packet size, and the number of packets, which are the current measurement results (for X seconds). FIG. 3A shows the flow packet size and the number of packets, which are cumulative measurement results from the occurrence of the flow to the present. FIG. 3B shows the queue length currently used by the communication device 9 and the maximum queue length that is the buffer size of the queue as information independent of the flow ID.

  In this way, by calculating the packet discard probability using information related to the size of the flow passing through the communication device 9, a flow having a large size and a high rate can be specified as a flow to be controlled. If the size of all packets in the flow is constant, instead of summing up the total packet size, an approximate value of the total packet size is obtained by multiplying the standard predetermined packet size by the number of packets. I want. As described above, by using the approximate value, the calculation amount of the information collecting unit 11 is reduced, so that high-speed processing can be expected.

  Information collection is realized by referring to header information of a flow packet. In addition, about the packet used as the object of the information to collect, you may collect from all the packets which arrived at the communication apparatus 9, or you may collect only the information of the packet sampled with the predetermined sampling rate from the arrived packets. Good. For example, a sampling rate of 0.1 means that one packet out of 10 passing packets is targeted for collection.

  To decide whether to discard a packet that has not been sampled, for example, when it arrives, only the flow ID is referenced from the header, and the packet is discarded using the latest value calculated for the probability of discarding the packet related to that flow. There is a method of deciding whether or not to do.

The discard probability calculation unit 12 receives the information collected in S11, updates the table information (see FIG. 3) storing them, and calculates the discard probability of the packet that arrives at the communication device 9 while referring to the table. (S12). The discard probability P (X) of the packets constituting the flow X is calculated as shown in Equation 1.
P (X) = {S (X, t)} × {A1 (X, t) × A2 (X, t) ×... × Ai (X, t)} (Equation 1)

  Here, S (X, t) represents the flow size at the time t of the flow X, and Ai (X, t) (i is a natural number) represents information other than the size at the time t of the flow X. Ai (X, t) includes, for example, a flow rate and a queue length of the communication device 9. Further, in Equation 1, P (X) is calculated by multiplying each parameter, but addition may be used instead of multiplication. When the calculated value of the discard probability P (X) is 1 or more, the discard probability is 1.

  The discard probability P (X) of Equation 1 may be changed according to a continuously increasing function based on these parameter values as shown in FIG. 5A, or a certain threshold is provided as shown in FIG. 5B. Thus, it may be changed according to the step function. FIG. 5 shows an example of how P (X) changes, and other changes may be made. In the actual communication device 9, the continuous increase function and the step function are functions such that the discard probability is always 1 when a flow size larger than a certain value arrives in consideration of the processing allowable amount of the communication device 9. .

  5 is 0 to 1, which is a relative value obtained by dividing the flow size for a predetermined flow by the flow sizes of all the flows of the communication device 9 through which the predetermined flow passes. . A value obtained by dividing the flow size of a predetermined flow by the maximum flow size of all flows may be used as a relative value. By using the relative value in this way, it is possible to identify a flow that is relatively troublesome. In addition, although it is desirable to use the relative value divided by the parameters relating to all flows, the parameter of Formula 1 such as the flow size may be a parameter value that is not divided.

  The packet discard unit 13 discards the packet that has arrived at the communication device 9 according to the packet discard probability calculated in S12 (S13). Packet discard may be performed randomly as shown in FIG. 6 (a) or periodically (discard the first packet → discard the second packet → three) as shown in FIG. 6 (b). The packet of the eye may be discarded. Note that the packet discarding unit 13 regards the packet discard probability calculated in S12 as being a control target flow or higher and controls the control target flow with priority over other flows. It is desirable.

  The above is the flow of the congestion control method. By this method, the congestion control considering the flow size in real time can be performed, so that it is possible to suppress the congestion by preferentially controlling the flow to be controlled that is likely to cause a spike causing the congestion.

  The present invention described above can be widely modified without departing from the spirit thereof as follows.

  For example, the program executed by the congestion control device 1 may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. . Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information such as a network such as the Internet or a communication line such as a telephone line. The program may be for realizing a part of the above-described processing. Furthermore, what can implement | achieve the above-mentioned process in combination with the program already recorded on another apparatus, what is called a difference file (difference program) may be sufficient.

  Further, as a second embodiment, instead of the first embodiment shown in FIG. 1 in which all necessary means for congestion control are performed by one communication device 9, a communication device 9 having an information addition device 1A as shown in FIG. , You may put it. The information addition apparatus 1A located on the upstream side of the flow has the information collection unit 11 of FIG. 2, and adds the information collected by the information collection unit 11 to the packet header, and the congestion control located on the downstream side of the flow Notify the discard probability calculation unit 12 of the device 1.

  The packet header for writing the collected information may be a ToS (Type of Service) field of the IP header, or a FlowLabel or MPLS label of an IPv6 (Internet Protocol Version 6) header. As a result, means necessary for congestion control are distributed to the communication devices 9 in the network, so that even in a network having a high-speed and large-capacity line that requires high-speed processing in the communication device 9, the processing catches up with the transfer rate, so Control can be performed.

  Furthermore, as a third embodiment, instead of the first embodiment shown in FIG. 1 and the second embodiment shown in FIG. 7, a congestion control device that centrally manages a plurality of communication devices 9 as shown in FIG. 1 may be used. 1 and 7 is a configuration in which the congestion control device 1 corresponds to the communication device 9 on a one-to-one basis.

  In the third embodiment, the congestion control device 1 includes an information collection unit 11 and a discard probability calculation unit 12. The communication device 9 includes a packet discard unit 13. As a result, the number of congestion control devices required in the network can be reduced, and the congestion control device 1 can be easily managed.

  Note that the control target flow controlled by the congestion control device is a set of flows having a large flow size in a flow size distribution obtained by accumulating the flow existence probabilities for each flow size as shown in FIG. 9, for example. For example, the congestion control device may set a flow whose flow size is equal to or larger than a predetermined value (500 [Mbits], etc.) as a control target flow, or a predetermined ratio (0.1%, etc.) from a flow having a larger flow size. The flow may be a control target flow.

It is a block diagram which shows the network system regarding 1st Embodiment of this invention. It is a block diagram which shows the congestion control apparatus regarding 1st Embodiment of this invention. It is explanatory drawing which shows the information for calculating the discard probability of the packet regarding 1st Embodiment of this invention. It is a flowchart which shows the operation | movement of congestion control regarding 1st Embodiment of this invention. It is explanatory drawing which shows the calculation method of the packet discard probability regarding 1st Embodiment of this invention. It is explanatory drawing which shows the discard method of the packet regarding 1st Embodiment of this invention. It is a block diagram which shows the network system regarding 2nd Embodiment of this invention. It is a block diagram which shows the network system regarding 3rd Embodiment of this invention. It is a graph which shows an example of the flow of the control object which the congestion control apparatus regarding each embodiment of this invention controls.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Congestion control apparatus 9 Communication apparatus 1A Information addition apparatus 11 Information collection part 12 Discard probability calculation part 13 Packet discard part 14 Control interface

Claims (9)

A congestion control method in which a communication device that relays a flow discards a packet of the flow to control congestion,
First communication apparatus located upstream of the flow of said communication device, for each active flow through the communication device, and stores in the storage means to collect the flow information including the flow size is accumulated, the By describing the flow size in the packet header of the flow, it notifies the second communication device located downstream of the first communication device,
The second communication device of the communication devices is
For each flow that passes through its own device, the notified flow size is read from the packet header, and a packet is determined from the ratio of the flow size of a predetermined flow that passes through the own device in the flow size of all flows that pass through the own device. A discard probability calculation procedure for calculating a discard probability of the packet of the predetermined flow by calculating a predetermined function for obtaining the discard probability of
A packet discarding procedure for discarding the packet of the active flow that is the predetermined flow according to the discard probability of the packet;
The congestion control method characterized by performing.   The congestion control method according to claim 1, wherein the information collection procedure calculates the flow size by calculating a total size of packets arriving at the communication apparatus.   The congestion control method according to claim 1, wherein the information collection procedure calculates the flow size by multiplying a total number of packets arriving at the communication apparatus by a predetermined packet size.   The congestion control method according to claim 1, wherein the information collection procedure collects information from all packets arriving at the communication apparatus for a predetermined flow.   The congestion control method according to claim 1, wherein the information collecting procedure collects information from a part of all packets arriving at the communication device for a predetermined flow. The congestion control method according to claim 1, wherein the discard probability calculation procedure obtains the discard probability of the packet according to a continuous increase function having the flow size ratio as a domain as the predetermined function . The congestion control method according to claim 1, wherein the discard probability calculation procedure obtains the discard probability of the packet according to a step function having the flow size ratio as a domain as the predetermined function . A congestion control program for causing a computer to execute the congestion control method according to any one of claims 1 to 7 . A congestion control system in which a communication device that relays a flow discards a packet of the flow to control congestion,
The congestion control system includes: a first communication device positioned upstream of a flow of the communication devices; and a second communication device of the communication devices positioned downstream of the first communication device. Are connected by a network,
The first communication device collects flow information including the accumulated flow size for each active flow passing through the communication device, stores the flow information in a storage unit, and describes the flow size in a packet header of the flow. To notify the second communication device,
The second communication device is
For each flow that passes through its own device, the notified flow size is read from the packet header, and a packet is determined from the ratio of the flow size of a predetermined flow that passes through the own device in the flow size of all flows that pass through the own device. A discard probability calculation procedure for calculating a discard probability of the packet of the predetermined flow by calculating a predetermined function for obtaining the discard probability of
A packet discarding procedure for discarding the packet of the active flow that is the predetermined flow according to the discard probability of the packet;
A congestion control system characterized by executing

Images