JP5284997B2 - Communication apparatus and communication method having TCP retransmission timeout value dynamic change function, and program therefor - Google Patents

Description

  The present invention relates to a communication quality control technique in a packet communication network such as the Internet or an intranet, and is particularly suitable for improving the throughput by optimizing the transmission rate of a data communication protocol such as TCP (Transmission Control Protocol). The present invention relates to a TCP parameter setting technique.

  In a packet communication network represented by the Internet, since a plurality of users share the network, the packet transfer quality depends on the use situation of the network.

  However, in recent years, network technology such as DiffServ (Differentiated Services) can guarantee packet transfer quality in units of flows (packets that have the same source and destination address information and port number). It is possible.

  As for the quality assurance (QoS: Quality of Services) technology using DiffServ, for example, there are those described in Patent Document 1 and Patent Document 2.

  In this technology, the policing rate and burst size are specified for the flow for quality assurance in the network, and the inflow restriction of traffic is realized at the edge router (device installed at the entrance of the network) for the traffic exceeding the specification. Policing (discarding excess packets) is performed. Commercial products that use a token bucket policer for policing are widely used (see FIG. 4 described later).

  In communication in such an environment, Patent Document 1 and Patent are applied to the flow for performing UDP (User Datagram Protocol) communication with a substantially constant transmission rate such as video distribution and voice call by appropriate control on the transmission side. The quality assurance technique described in Document 2 enables transfer without packet loss.

  However, in communications with congestion control (window control) functions such as TCP that gradually increase the transmission rate and decrease the transmission rate at a certain rate when packet loss occurs, packet loss has a large effect on throughput. .

  For example, in a network that performs quality assurance as described in Patent Documents 1 and 2, a problem has been disclosed in which the throughput of TCP is significantly lower than a prescribed rate due to policing.

  As a technique for solving this problem, there is a shaping technique for smoothing the transmission rate of packets as described in Patent Document 3, for example. In this shaping technique, traffic transmitted from a terminal is passed through a shaping device, and the traffic transmission rate is smoothed within the shaping device so as to be within the network traffic regulations.

  However, since the shaping device needs to be installed between the network and the transmission terminal, each user who subscribes to the service needs to have the shaping device. However, since a shaping apparatus is expensive, the use of a quality assurance network based on the shaping apparatus is not practical from the viewpoint of cost.

JP 2004-320489 A JP 2005-073106 A Japanese Patent Laid-Open No. 06-268671

  The problem to be solved by the present invention is that, in a quality assurance network using a conventional technology, for example, a network technology such as DiffServ, traffic exceeding a secured rate is discarded by policing of an edge router. In a communication protocol that reduces the transmission rate each time a packet loss occurs, it has been necessary to use the above-described shaping device in order to prevent a reduction in throughput due to policing. However, a service subscriber prepares a shaping device, In an environment that uses existing token bucket policers (policing using token buckets) widely implemented in commercial products, in terms of the difficulty in cost and the approach to improve the policing mechanism itself, Policing mechanism for the entire network Te there is a need to rebuild replaced by a new device is that it become a major cost burden for operators.

  Accordingly, an object of the present invention is to provide a communication apparatus and a communication method that can improve the throughput by solving these problems of the prior art and optimally setting a retransmission timeout value that is a parameter of TCP communication, and a program therefor It is to be.

In order to achieve the above object, the present invention adopts the following configuration.
a) When performing TCP communication, the communication device according to the present invention manages the same transmission / reception address information and port information as one flow in the transmission side environment, detects a retransmission timeout for each flow, and retransmits the timeout. Throughput measurement recording means for measuring and recording the throughput until the occurrence of the occurrence, changing the retransmission timeout value, and obtaining the retransmission timeout value that maximizes the throughput by measuring the throughput by the throughput measurement recording means It is characterized by having a retransmission timeout value setting means set as a new retransmission timeout value and a communication means for performing communication based on the set maximum retransmission timeout value.

b) In the communication device described in a) above, the retransmission timeout value is changed within a range equal to or longer than the round-trip delay time.

c) In the communication device described in a) above, a means for obtaining a policing setting value set in a network to which the token bucket policer is applied, and a transmission terminal sent from the time of the first retransmission timeout occurrence Based on the packet information, the measuring means for measuring the time until the recovered transmission rate exceeds the policing rate, and the time obtained by the measuring means is subtracted from the time when the bucket size is satisfied from 0 to the upper limit value. And a means for setting the value as a new retransmission timeout value when the value is larger than the round-trip delay time.

d) In the communication device described in a) above, the retransmission timeout value setting means uses a binary search method in which the retransmission timeout value is set to the initial value of the current retransmission timeout value when the retransmission timeout occurs. It is characterized by means for temporarily changing, searching for a retransmission timeout value that maximizes throughput, and setting the searched maximum retransmission timeout value as a new retransmission timeout value.

e) In the communication apparatus described in d) above, the search termination condition of the binary search method is that the throughput is not improved or the throughput is a value set as a target for setting the policing rate. Alternatively, the change width of the retransmission timeout is either smaller than a predetermined threshold value.

f) In the communication device described in a) above, the change of the retransmission timeout value is performed in advance in the range from the round-trip delay time to the generally set retransmission timeout value, triggered by the occurrence of the retransmission timeout. It is characterized by being performed at a predetermined arbitrary interval.

g) A program according to the present invention is a program that causes a computer to function as each unit in the communication device according to any one of a) to f).

h) In the communication method according to the present invention, when TCP communication is performed, the same transmission / reception address information and port information are managed as one flow in the transmission side environment, and a retransmission timeout is detected for each flow. A throughput measurement recording procedure for measuring and recording the throughput until the occurrence of the error, and changing the retransmission timeout value, and obtaining a retransmission timeout value that maximizes the throughput by measuring the throughput by the throughput measurement recording procedure. It is characterized by having a retransmission timeout value setting procedure set as a new retransmission timeout value and a communication start procedure for performing communication based on the set maximum retransmission timeout value.

i) In addition, in the communication method described in h) above, the sending terminal sends out the procedure for obtaining the policing setting value set in the network to which the token bucket policer is applied and the first retransmission timeout. Based on the packet information, the measurement procedure for measuring the time until the recovered transmission rate exceeds the policing rate, and the time obtained by the measurement means are subtracted from the time when the bucket size is satisfied from 0 to the upper limit value. And a procedure for setting the value as a new retransmission timeout value when the value is larger than the round trip delay time.

j) In the communication method described in h) above, the retransmission timeout value setting procedure uses a binary search method in which a retransmission timeout value is used as an initial value when the retransmission timeout occurs. The procedure is characterized in that it is a procedure for temporarily changing, searching for a retransmission timeout value that maximizes throughput, and setting the searched maximized retransmission timeout value as a new retransmission timeout value.

  According to the present invention, for example, when TCP communication is performed in a network to which policing by a token bucket is applied, the retransmission timeout value is dynamically set to a retransmission timeout value that maximizes the throughput. Communication that can minimize the non-transmission time during which transmission is not performed even though the token is full, and as a result, the token can be used effectively and the average throughput can be made closer to the policing rate. An apparatus, a communication method, and a program therefor can be provided.

It is a figure which shows the relationship between time and token amount about the waste of time until resending. It is a figure which shows <initialization setting subroutine> common to all the Examples. It is a figure which shows the <throughput calculation subroutine> common to all the Examples. It is a block diagram which shows the environment which concerns on this invention. It is a figure which shows the flowchart of the process in the transmission side environment which concerns on Example 1 of this invention. It is a figure which shows the flowchart of the process in the transmission side environment which concerns on Example 2 of this invention. FIG. 10 is a diagram illustrating changes in parameters for each occurrence of retransmission timeout in the second embodiment of the present invention. It is a figure which shows the flowchart of the process in the transmission side environment which concerns on Example 3 of this invention. It is a figure which shows the flowchart of the process in the transmission side environment which concerns on Example 3 of this invention (when the search end condition is that the change width of RTO by the binary search method became smaller than the threshold value). It is a figure which shows the change of each parameter for every retransmission timeout generation in Example 3 of this invention. It is a figure which shows the flowchart of the process in the transmission side environment which concerns on Example 4 of this invention. It is a figure which shows the change of each parameter for every retransmission timeout generation in Example 4 of this invention. It is a figure which shows the flowchart of the process in the transmission side environment which concerns on. It is a figure which compares the throughput in the retransmission timeout calculated | required in Examples 2-4 which concern on this invention, and an initial retransmission timeout.

<Points of the present invention>
The present invention is a technique in which a retransmission timeout value that maximizes throughput is obtained by dynamically changing a retransmission timeout value, and communication is performed using a retransmission timeout value that maximizes the throughput. Since the present invention is particularly suitable when applied to a network environment using a token bucket policer widely used in products, a case where the token bucket policer is used for a network using a token bucket policer will be described below as an example.

  Conventionally, for example, when performing TCP communication in a network to which policing by a token bucket is applied, a retransmission timeout minimum value, which is a TCP parameter in a transmission side environment, is fixedly allocated, and the effect of token efficiency This was a factor that hindered use, and the average throughput was greatly reduced compared to the polishing rate.

  In the present invention, by selecting and setting the retransmission timeout value so that the throughput is maximized in the transmission side environment, the token can be efficiently used when the retransmission timeout occurs, and as a result, the average throughput is reduced. It is possible to make it closer to the policing rate.

<Explanation of terms used in the present invention and technical significance of the present invention>
Policing is a method of monitoring the flow rate for each flow and discarding packets exceeding a certain transmission rate without performing buffering. Unlike shaping, policing is characterized in that no delay occurs because queuing (buffering) of packets is not performed.

  A token bucket is a control mechanism that directs when to forward traffic based on the presence of tokens in the bucket. The token in the bucket is deleted when sending the packet.

  A maximum value (maximum bucket size Bs) is set in the token bucket, and the token is not replenished beyond the maximum bucket size Bs. The tokens are replenished according to the policing rate Bp for the maximum bucket size Bs.

  The time required for the token to be replenished from the zero state to the maximum bucket size Bs is defined as a rate coefficient Br. When the packet arrives, the token amount in the bucket is confirmed, and when there is a token, the packet can be transferred. However, when the token is exhausted, the packet is not transferred but discarded.

  In the present invention, policing using a token bucket is defined as a “token bucket policer”.

  In a network under the application of a token bucket policer, in TCP communication, burst traffic is transmitted in order to increase the transmission rate when receiving a ACK until the transmission side detects packet loss by autonomous window control.

  Therefore, when the transmission rate exceeds the policing rate Bp, packet discard occurs. In addition, the detection of the packet loss is also triggered by the arrival of the ACK packet.

  On the other hand, since there is a time difference between round-trip delay time between packet discard and ACK arrival time, the congestion window remains large even immediately after the policer discards the packet, and packets are sent one after another even if there is no surplus token in the policer As a result, burst discard occurs.

  When a packet is discarded in this way, TCP tries to retransmit the discarded packet, but the upper limit of the number of retransmission attempts depends on the size of the congestion window at the timing when the packet is discarded. Therefore, when the number of re-transmission attempts is small, there is a high possibility that the packet re-transfer will fail and a retransmission timeout will occur depending on the number of discarded packets.

  The retransmission time-out value RTO has a fixed minimum value RTO_MIN (generally set to a value of 200 msec or more). When a retransmission time-out occurs, a waiting time of RTO_MIN or more is always generated.

  When a retransmission timeout occurs due to burst discard due to token depletion, if RTO_MIN is larger than the rate coefficient Br, useless waiting time that communication is not resumed even though the token is satisfied up to the maximum bucket size Bs ( RTO_MIN-Br) always occurs.

  This waiting time is a factor that greatly reduces the throughput with respect to the expected policing rate Bp. FIG. 1 shows the relationship between the time and the token amount with respect to the waste of time until the above-described retransmission.

  In the present invention, when a retransmission timeout occurs due to burst discard due to token depletion in TCP communication under policing application, the retransmission timeout value RTO is allowed to be smaller than RTO_MIN, and the retransmission timeout value RTO is temporarily set. The retransmission timeout value RTO that maximizes the throughput is obtained by recording and comparing the throughput at that time, and a throughput closer to the policing rate than the conventional one is realized. If the retransmission time-out value RTO is too small, useless retransmissions occur. Therefore, it is necessary to select a suitable range for the change of the retransmission time-out value RTO.

<Description of Examples>
Hereinafter, embodiments according to the present invention will be described in detail with reference to flowcharts. Note that common processing is prepared in advance as a subroutine for all the embodiments. Common processes here include the <initialization setting subroutine> and the <throughput calculation subroutine> described below. First, these subroutines will be described.

<Initialization subroutine>
In each embodiment, in order to obtain the retransmission timeout value RTO for maximizing the throughput, RTO_max and Throughput_max are defined as variables for storing the information, and a flow for setting the initial values is defined as <initialization setting subroutine>.

FIG. 2 is a flowchart of <Initialization setting subroutine>.
As shown in the figure, first, a generally set retransmission timeout value RTO (≈RTT + RTO_MIN) is defined as RTO_INIT, and is set as an initial value of the retransmission timeout value RTO (step S001). Here, RTT (Round Trip Time) is a round trip delay time.

  Next, the transmission side terminal detects that a retransmission timeout has occurred, and records the occurrence time t1 (step S002). The amount of packets transmitted from the start of communication until t1 is obtained from the transmission packet information and divided by the elapsed time from the start of communication to t1, thereby defining and calculating the throughput as a variable Throughput (step) S003).

  RTO_INIT set in step S001 is set to RTO_max, and Throughput obtained in step S003 is set to Throughput_max (step S004).

<Throughput calculation subroutine>
In each embodiment, in order to obtain the retransmission timeout value RTO that maximizes the throughput, the retransmission timeout value RTO is set with a different algorithm (see Embodiments 2 to 4). Every time a retransmission timeout occurs, the retransmission timeout value RTO is changed and then the throughput is calculated. The flow for calculating the throughput is defined as a throughput calculation subroutine.

FIG. 3 is a flowchart of <throughput calculation subroutine>.
As shown in the figure, first, assuming that the number of occurrences of retransmission timeout is n (n ≧ 2), the transmission side terminal detects the nth retransmission timeout and records the occurrence time tn (step S005).

  Using the previous (n−1) th retransmission timeout occurrence time tn−1, the packet amount transmitted from tn−1 to tn is acquired from the transmission packet information, and the acquired packet amount is acquired from tn−1 to tn. The throughput is calculated as a variable Throughput by dividing by the time (step S006).

  Throughput and Throughput_max obtained in step S006 are compared (step S007). If Throughput is greater than Throughput_max (step S007: Y), RTO_max and Throughput_max are updated (step S008). If Throughput is smaller than Throughput_max (step S007: N), the throughput calculation process is terminated without performing anything.

<Example 1>
Next, Example 1 will be described.
In the first embodiment, the transmitting terminal is connected to the edge router of the network through an access line. This edge router is subjected to policing by token buckets defined by policing setting values (policing rate Bp, maximum bucket size Bs, rate coefficient Br).

FIG. 4 is a diagram showing the above environment, and is a common configuration in the first to fourth embodiments of the present invention.
In the figure, the transmission terminal 10 performs TCP communication from the access line 20 and the edge router 31 to the reception terminal 50 via the DiffServ network 30, the edge router 32, and the access line 40.

  The transmission terminal 10 notifies the Diffserv network 30 of the flow transmission rate before the start of communication to ensure the packet transfer quality.

  The edge router 31 manages the same address information and port information on the transmission side (transmission terminal) and reception side (reception terminal) as a flow, manages the transmission rate for each flow, and manages the flow rate exceeding the policing rate Bp. Discard.

  In the present embodiment, the occurrence of a retransmission timeout is detected, the throughput is measured based on the information of the transmitted packet every time the retransmission timeout occurs, and RTT ≦ RTO is triggered by the occurrence of the retransmission timeout during TCP protocol communication. In this range, a retransmission timeout value RTO that maximizes the throughput is searched, the value is set as a new retransmission timeout value RTO, and communication is performed.

FIG. 5 is a flowchart in the transmission side environment (transmission terminal 10) of the present embodiment.
In the figure, first, TCP communication is started, and the throughput for RTO_INIT is set as the initial value of Thoughtput_max in the initial value setting subroutine (the process of the flowchart shown in FIG. 2) (step S009).

  In this embodiment, the retransmission timeout value RTO is changed and a new throughput calculation is repeated using the retransmission timeout as a trigger (trigger), and therefore, the RTO at the nth retransmission timeout occurrence is defined as a variable RTO_n.

  The number of retransmission timeout occurrences n is set to 2 (step S010), RTO_n is calculated by a specific algorithm (see Examples 2 to 4 to be described later) (step S011), and the calculated RTO_n is set as a retransmission timeout value. Set as RTO (step S012) and perform communication.

  Next, the value of the round trip delay time RTT is acquired from the received ACK packet information (step S013). If the retransmission timeout value RTO set in step S012 is smaller than the round trip delay time RTT, useless retransmission timeout increases. These are compared (step S014), and if the retransmission timeout value RTO is greater than or equal to the round trip delay time RTT, that is, if RTT ≦ RTO (step S014: Y), communication is continued using the retransmission timeout value RTO set in step S012. (Step S015) Using the occurrence of a retransmission timeout as a trigger, Throughput is calculated in the throughput calculation subroutine (the process of the flowchart shown in FIG. 3), RTO_max and Thoughtput_max are updated (Step S016), and the number of retransmission timeout occurrences n Increase by one By (step S017), the flow returns to step S011.

  The processing from step S011 to step S017 is repeated, and the termination condition is that the retransmission timeout value RTO is smaller than the round trip delay time RTT. In this case, step S014: N), RTO_max that maximizes the throughput is set as the retransmission timeout value RTO. Is set (step S018), and communication is continued.

  Next, regarding the retransmission timeout value RTO search algorithm that maximizes the throughput in step S011, three methods from Example 2 to Example 4 are proposed, and a flowchart and a specific example are shown for each method.

<Example 2>
In the second embodiment, in the environment of the first embodiment, when the transmission side (transmission terminal 10) can acquire information on the policing setting values (policing rate Bp, maximum bucket size Bs, rate coefficient Br) during communication of the TCP protocol, The time T from when the first retransmission timeout occurs until the throughput exceeds the policing rate Bp is measured based on the packet information transmitted from the transmission terminal 10, and the difference between the rate coefficient Br and the time T is set as the retransmission timeout value RTO. (RTO = Br−T), the throughput in the case of this retransmission timeout value RTO is calculated, the newly calculated throughput is compared with the previous throughput, and when the throughput is improved, the throughput is improved to the new retransmission timeout value RTO. And performing communication.

  When a retransmission timeout occurs, in order to minimize the transmission rate, the sufficient amount exceeds the token consumption until the transmission rate exceeds the policing rate Bp. Accordingly, the retransmission timeout value RTO is considered to be the optimum value obtained by subtracting the time T when the transmission rate exceeds the policing rate Bp with respect to the rate coefficient Br.

  If RTO <RTT is satisfied after the change, extra retransmissions occur. In this case, throughputs are compared by setting the round trip delay time RTT and the value of the rate coefficient Br as the retransmission timeout value RTO.

Next, the process of the said Example 2 is demonstrated using a flowchart.
FIG. 6 is a flowchart in the transmission side environment of the present embodiment.
In the figure, first, TCP communication is started, and the transmission side terminal (transmission terminal 10) acquires the policing setting values (policing rate Bp, maximum bucket size Bs, rate coefficient Br) (step S19), and then the initial Through the value setting subroutine (the process of the flowchart shown in FIG. 2), the throughput for RTO_INIT is set as the initial value of Throughput_max (step S020).

  Next, the time T from when the initial retransmission timeout occurs t1 until Throughput exceeds the policing rate Bp is measured based on the transmission packet information (step S021).

  Next, the retransmission timeout count n is set to 2 (step S022), the value of the round trip delay time RTT is acquired from the received ACK packet information (step S023), and the difference between the rate coefficient Br and the time T (Br-T ) And the value of the round trip delay time RTT (step S024). If the difference (Br−T) between the rate coefficient Br and the time T is larger than the round trip delay time RTT (step S024: Y), RTO_n is set to Br−. Calculated as T (step S025).

  On the other hand, when the difference (Br−T) between the rate coefficient Br and the time T is equal to or shorter than the round trip delay time RTT (step S024: N), a wasteful retransmission timeout occurs, so the rate coefficient Br and the round trip delay time. RTTs are compared (step S027). If the rate coefficient Br is greater than the round trip delay time RTT (step S027: Y), RTO_n is calculated as the rate coefficient Br (step S028), and the round trip delay time RTT is greater than or equal to the rate coefficient Br. In the case of (step S027: N), RTO_n is calculated as the round trip delay time RTT (step S029).

  RTO_n obtained in steps S025, S028, and S029 is set as a retransmission timeout value RTO (step S026), and communication is continued (step S030).

  Next, using the occurrence of a retransmission timeout as a trigger, Throughput is obtained by a throughput calculation subroutine (the process of the flowchart shown in FIG. 3), and RTO_max and Throughput_max are updated (step S031).

  Next, RTO_max that maximizes the throughput is set to the retransmission timeout value RTO (step S032), and communication is continued.

(Specific example of Example 2)
Next, a specific example of the second embodiment will be described.
In the specific example of the second embodiment, similarly to the first embodiment, the round-trip delay time RTT is set to 5 msec, the policing rate Bp is set to 2 Mbps, and the maximum bucket size Bs is set to 25 KB (value at which the rate coefficient Br is 100 msec) RTO_MIN that is set to 200 msec.

  In step S019 of FIG. 6, the policing setting values are obtained as the policing rate Bp of 2 Mbps, the maximum bucket size Bs of 25 KB, and the rate coefficient Br of 100 msec. In the initial value setting step of step S020, RTO_INIT is 205 msec. When the first retransmission timeout occurs, the throughput is calculated, Throughput is 1.16 Mbps, and these two values are set as RTO_max and Throughput_max.

  In step S021, a time T from when the first retransmission timeout occurs until Throughput becomes greater than the polishing rate Bp is measured. Time T is measured as 5 msec. In step S022, the number n of retransmission timeout occurrences is set to 2. In step S023, the round-trip delay time RTT value (5 msec) is obtained. In step S024, the difference between the rate coefficient Br and time T (Br-T). And the round-trip delay time RTT are compared.

  Since the difference (Br−T) between the rate coefficient Br and time T is 95 msec and is larger than the round trip delay time RTT 5 msec (step S024: Y), the retransmission timeout value RTO_n is obtained as 95 msec in step S025, and step S026. In step S030, the RTO_n obtained for the retransmission timeout value RTO is set, and communication is continued in step S030. In step S031, throughput is calculated with the occurrence of retransmission timeout as a trigger, and Throughput is calculated as 1.85 Mbps.

  Next, since Throughput is larger than Throughput (1.85 Mbps) and Throughput_max (1.16 Mbps), RTO_max is updated to 95 msec and Throughput_max is updated to 1.85 Mbps. In step S032, the retransmission timeout value RTO is set to 95 msec, and communication is continued.

  FIG. 7 is a diagram illustrating a specific example of the retransmission timeout values RTO, Throughput, RTO_max, and Thoughtput_max in the processing from step S19 to step S31.

  It can be seen that the throughput is 1.85 Mbps with respect to the determined retransmission timeout value RTO, which is an improvement of 0.69 Mbps compared to RTO_INIT (≈RTT + RTO_MIN = 205 msec). Also, the number of retransmission timeout occurrences n necessary for calculating this parameter is two.

<Example 3>
In the environment of the first embodiment, the third embodiment is triggered by the occurrence of a retransmission timeout during TCP protocol communication.
RTT ≤ RTO ≤ RTO_INIT
In this range, the retransmission timeout value RTO search range is defined as variables RTO_low to RTO_high and set as the round trip delay time RTT and the initial retransmission timeout value RTO_INIT, respectively, and the retransmission timeout value RTO that maximizes the throughput is binary search method. Search using.

  The search end condition may be that the throughput is not improved by the binary search, may be specified by the ratio of the policing rate aiming for the throughput, or the case where the RTO width to be changed is equal to or less than the round trip delay time RTT. Also good.

Next, the process of the said Example 3 is demonstrated using a flowchart.
8 and 9 are flowcharts of processing performed in the transmission-side environment according to the third embodiment. FIG. 8 is a flowchart when the search end condition is “the throughput is no longer improved”, and FIG. 9 is a flowchart when the search end condition is “the RTO width to be changed is equal to or less than the round trip delay time RTT”. is there. 8 and 9 are the same flowcharts except that the search end condition based on the throughput in step SS52 in FIG. 8 is replaced with the search end condition based on the retransmission timeout change width in step S055 in FIG.

  In FIG. 8, first, TCP communication is started, and the throughput for RTO_INIT is set as the initial value of Throughput_max in the initial value setting subroutine (the process of the flowchart shown in FIG. 2) (step S033).

  Next, in the search for the retransmission timeout value RTO that maximizes the throughput, the retransmission timeout value RTO search range is defined as variables RTO_low and RTO_high, and the respective throughputs are defined as Throughput_low and Throughput_high, and the retransmission timeout value obtained by the initial value setting subroutine RTO and Throughput are set to RTO_high and Throughput_high (step S034).

  Next, the value of the round trip delay time RTT is acquired from the received ACK packet information (step S035), the retransmission timeout value RTO is set as the round trip delay time RTT (step S036), and the retransmission timeout count n is set to 2 ( In step S037), through the occurrence of retransmission timeout, Throughput is obtained in the throughput calculation subroutine, and RTO_max and Throughput_max are updated (step S038).

  The retransmission timeout value RTO and the obtained Throughput are set to RTO_low and Througput_low, respectively (step S039), and a binary search is performed.

  The retransmission timeout occurrence number n is set to 3 (step S040), the throughput_low and the throughput_high are compared (step S041), and if the throughput_low is higher than the throughput_high (step S041: Y), a retransmission timeout value that maximizes the desired throughput. RTO is determined to be in the vicinity of RTO_low, RTO_high is updated to (RTO_low + RTO_high) / 2 (step S042), this value is set to the retransmission timeout value RTO (step S043), and communication is continued (step S044). .

  Next, using the occurrence of a retransmission timeout as a trigger, Throughput is obtained by a throughput calculation subroutine (the process of the flowchart shown in FIG. 3), and RTO_max and Throughput_max are updated (step S045).

  Then, Throughput_high is updated with the obtained Throughput (step S046).

  On the other hand, as a result of comparing Throughput_low and Throughput_high in Step S041, if Throughput_high is higher than Throughput_low (Step S041: N), it is determined that the retransmission timeout value RTO for maximizing the desired throughput is in the vicinity of RTO_high, and RTO_low is set to ( RTO_low + RTO_high) / 2 (step S047), this value is set as the retransmission timeout value RTO (step S048), and communication is continued (step S049).

  Next, with the occurrence of a retransmission timeout as a trigger, Throughput is obtained in a throughput calculation subroutine (the process of the flowchart shown in FIG. 3), and RTO_max and Throughput_max are updated (step S050). Then, Throughput_low is updated with the obtained Throughput (step S051).

  Steps S042 to S046 and steps S047 to S051 are set to RTO_high and RTO_low, respectively, to determine whether Throughput_max has been updated (Step S052). When Throughput and Throughput_max are equal, that is, Throughput_max is updated (Step S052: N) The retransmission timeout count n is increased by 1 (step S053), and the processing from steps S041 to S053 is repeated.

  In FIG. 8, as described above, the repetition termination condition is “the maximum throughput value Througput_max is not updated due to the retransmission timeout value RTO change by binary search” (step S052). If Throuput_max is not updated, that is, if Throughput and Throughput_max are not equal (step S052: Y), RTO_max is set to the retransmission timeout value RTO (step S054), and communication is continued.

  FIG. 9 shows an example in which, as described above, the iterative termination condition is “the change width (RTO_low + RTO_high) / 2 of the retransmission timeout value RTO in binary search is smaller than the threshold σ” (step S055). It is. When the change width (RTO_low + RTO_high) / 2 of the retransmission timeout value RTO is smaller than the threshold σ (step S055: Y), RTO_max is set to the retransmission timeout value RTO (step S054), and communication is continued.

(Specific example of Example 3)
Next, a specific example of Example 3 will be described.
As a specific example of the third embodiment, the round-trip delay time RTT is 5 msec (RTO_low = RTT = 5 msec), the generally set RTO_MIN is 200 msec, RTO_high is RTO_INIT (≈RTT + RTO_MIN = 205 msec), and the polishing rate Bp is 2 Mbps. The maximum bucket size Bs is 25 KB (value at which the rate coefficient Br is 100 msec), and the search end condition is that the throughput is not improved.

  In step S033, RTO_INIT is set to 205 msec, the throughput is calculated when the first retransmission timeout occurs, Throughput is 1.16 Mbps, and these two values are set to RTO_max and Thoughtput_max. In step S034, 205 msec is set for RTO_high and 1.16 Mbps is set for Thoughtput_high.

  The round-trip delay time RTT value (5 msec) is acquired in step S035 in FIG. 8, and a new retransmission timeout value RTO is set to the RTT value 5 msec in step S036.

  Next, in step S037, the number of retransmission timeouts is set to 2, and in step S038, the throughput is calculated with the occurrence of retransmission timeout as a trigger, and the Thoughtput is calculated to be 1.69 Mbps.

  Next, Throughput (1.69 Mbps) and Throughput_max (1.16 Mbps) are compared. Since Throughput is large, RTO_max is updated to 5 msec, and Thoughtput_max is updated to the value of Thoughtput at this time. In step S039, RTO_low is set to 5 msec, and Throughput_low is set to 1.69 Mbps.

  In step S040, the retransmission timeout occurrence number n is set to 3, and in step S041, Throughput_low and Throughput_high are compared. In this example, Throughput_low (1.69 Mbps) is larger than Throughput_high (1.16 Mbps) (step S041: Y), RTO_high is calculated as 105 msec in step S042, and RTO_high is set as a retransmission timeout value in step S043. RTO is set and communication is continued in step S044.

  Throughput is calculated with the occurrence of retransmission timeout as a trigger, and Throughput is calculated as 1.9 Mbps (step S045). Compared to Throughput and Throughput_max, since Throughput is greater than Throughput_max, RTO_max is updated to 105 msec and Throughput_max is updated to 1.9 Mbps.

  In Step S046, Throughput_high is updated to 1.9 Mbps. In Step S052, Throughput and Throughput_max are compared. If Throughput and Throughput_max are equal (Step S052: N), this means that the Throughput_max has been updated. Therefore, the number of retransmission timeouts is updated to 4 in step S053, and the processing from steps S041 to S053 is repeated again.

  FIG. 10 is a diagram illustrating examples of values of retransmission timeout values RTO, Throughput, RTO_max, Thoughtput_max, RTO_high, Throughput_high, RTO_low, and Throughput_low in the iterative processing (n = 1 to 5) from step S041 to S053 of the third embodiment. is there.

  If Throughput and Throughput_max are not equal in step S052 (step S052: Y), this means that Throughput_max is no longer updated, and thus the iterative process from steps S041 to S053 is terminated, and in step S054. The RTO_max of 55 msec is set as the retransmission timeout value RTO, and communication is continued.

  It can be seen that the throughput is 1.93 Mbps with respect to the determined retransmission timeout value RTO, which is 0.77 Mbps higher than RTO_INIT (≈RTT + RTO_MIN = 205 msec).

  Further, the number of retransmission timeout occurrences necessary for calculating this parameter is five.

<Example 4>
In the environment of the first embodiment, the fourth embodiment is triggered by the occurrence of a retransmission timeout during TCP protocol communication.
RTT ≤ RTO ≤ RTO_INIT
The RTO is sequentially changed at an arbitrary interval d within a range satisfying the above, exhaustively searching for an RTO that maximizes the throughput, and setting the value as a new retransmission timeout value RTO for communication.

Next, the process of the said Example 4 is demonstrated using a flowchart.
FIG. 11 is a flowchart of processing performed in the transmission-side environment according to the fourth embodiment.
As shown in the figure, first, TCP communication is started, and the throughput for RTO_INIT is set as an initial value of Throughput_max in an initial setting subroutine (the process of the flowchart of FIG. 2) (step S056).

  Next, the retransmission timeout number n is set to 2 (step S057), RTO_n at the n-th retransmission timeout is subtracted from the retransmission timeout value RTO by a fixed interval d (step S058), and RTO_n is set to the retransmission timeout value RTO (step S057). Step S059).

  Next, the value of the round trip delay time RTT is acquired from the received ACK packet information (step S060), and the round trip delay time RTT and the retransmission timeout value RTO are compared (step S061).

  As a result of the comparison in step S061, if the retransmission timeout value RTO is larger than the round trip delay time RTT (step S061: Y), communication is continued (step S062), and a throughput calculation subroutine (FIG. Throughput in step 3) is performed, and RTO_max and Throughput_max are updated (step S063).

  Next, the number of retransmission timeouts is increased by 1 (step S064), and the processing from steps S057 to S063 is repeated.

  As a result of the comparison in step S061, if the retransmission timeout value RTO is equal to or shorter than the round trip delay time RTT (step S061: N), a useless retransmission timeout occurs, so the search is stopped and RTO_max is set to the retransmission timeout value RTO. (Step S065) and communication is continued.

(Specific example of Example 4)
Next, a specific example of the fourth embodiment will be described.
As a specific example for the fourth embodiment, the round-trip delay time RTT = 5 msec, the polishing rate Bp is 2 Mbps, the maximum bucket size Bs is 25 KB (value at which the rate coefficient Br is 100 msec), d = 10 msec, and generally set RTO_MIN Is set to 200 msec.

  In the initial value setting subroutine of step S056, RTO_INIT is set to 205 msec, the throughput is calculated when the first retransmission timeout occurs, Throughput is 1.16 Mbps, and these two values are set to RTO_max and Throughput_max.

  In step S057, the retransmission timeout number n is set to 2, and in step S058, RTO_2 is set as RTO-d = 205-10 = 195 msec. In step S059, this value is set as a new retransmission timeout value RTO. In S060, the value of the round trip delay time RTT (5 msec) is acquired, and in step S061, the round trip delay time RTT is compared with the retransmission timeout value RTO.

  As a result of comparison between the round trip delay time RTT and the retransmission timeout value RTO in step S061, the retransmission timeout value RTO (195 msec) is larger than the round trip delay time RTT (5 msec) (step S061: Y), so that communication is performed in step S062. In step S063, 1.21 Mbps is calculated for Throughput by calculating throughput with the occurrence of retransmission timeout as a trigger. Since Throughput is larger than Throughput_max, RTO_max is updated to 195 msec and Throughput_max is updated to 1.21 Mbps.

  Next, in step S064, the number of retransmission timeouts n is increased by 1 and set to 3, and the processing from steps S058 to S064 is repeated.

  FIG. 12 shows values of retransmission timeout values RTO, Throughput, RTO_max, and Thoughtput_max in the iterative processing from steps S058 to S064.

  When RTO_n is decreased by d (= 10 msec) in step S058 and the termination condition in step S061 (round trip delay time RTT is greater than or equal to the retransmission timeout value RTO) (step S061: N), the throughput is maximized in step S065. The RTO_max of 75 msec is set as a retransmission timeout value RTO, and communication is continued.

  It can be seen that the throughput is 1.96 Mbps with respect to the determined retransmission timeout value RTO (75 msec), and the throughput is improved by 0.8 Mbps compared to RTO_INIT (≈RTT + RTO_MIN = 205 msec). Further, as can be seen from FIG. 12, in this example, the number of retransmission timeout occurrences n necessary for calculating the parameter is 20 times.

<Comparison of Throughput when Examples 2 to 4 are used>
Next, comparison of results obtained in Examples 2 to 4 described later will be described with reference to the drawings.
FIG. 13 is a diagram illustrating a comparison result between the retransmission timeout value RTO and the throughput of RTO_INIT obtained using the algorithm described in the second to fourth embodiments. FIG. 8A shows a table of numerical examples for each of the embodiments described above. FIG. 6B shows the value of the retransmission timeout value RTO obtained in each embodiment on the horizontal axis. , RTO_INIT throughput values are plotted on a vertical axis.

  According to FIG. 13, the retransmission timeout value RTO obtained by using the algorithm of the embodiments 2 to 4 is compared with the throughput when the algorithm of the embodiments 2 to 4 is not used (that is, when the retransmission timeout value RTO_INIT is set). It can be seen that the throughput is significantly improved when is set.

  The initialization setting subroutine shown in FIG. 2, the throughput calculation subroutine shown in FIG. 3, the flowchart of the first embodiment shown in FIG. 5, the flowchart of the second embodiment shown in FIG. The processing of the flowchart of the third embodiment shown in FIG. 9 and the processing of the flowchart of the fourth embodiment shown in FIG. 11 are a CPU, a memory, and a register built in the computer that constitutes the transmission-side environment (mainly a communication device serving as a transmission terminal). It implement | achieves by executing the program corresponding to each process using hardware, such as. A program corresponding to each of these processes can be distributed to the market via a recording medium such as FD, CD-ROM, or DVD, or a network such as the Internet. In addition, each step (each procedure) in the processing of each flowchart described above is processed using hardware such as a CPU, a memory, and a register built in a computer that constitutes a communication device that is a transmission side environment. “Means” in the claims means a component for executing these processes (steps).

10: Transmission terminal (transmission side communication device)
20: Access line 30: DiffServ network 31, 32: Edge router 311: Token bucket policer 33: Policer controller 40: Access line 50: Receiving terminal (reception side communication device)
Bp: Polishing rate Bs: Maximum bucket size Br: Rate coefficient

Claims (10)

When TCP communication is performed, the same transmission / reception address information and port information are managed as one flow in the transmission side environment, and the retransmission timeout is detected for each flow and the throughput until the retransmission timeout occurs is measured. Throughput measurement recording means for recording;
A retransmission timeout value setting means for changing a retransmission timeout value, obtaining a retransmission timeout value that maximizes the throughput by measuring the throughput by the throughput measurement recording means, and setting it as a new retransmission timeout value;
A communication unit configured to perform communication based on the set maximized retransmission timeout value. The communication device according to claim 1.
The communication apparatus is characterized in that the retransmission timeout value is changed within a range equal to or longer than the round trip delay time. The communication device according to claim 1,
Means for obtaining a policing setting value set in a network to which a token bucket policer is applied;
Measuring means for measuring the time from when the first retransmission timeout occurs until the recovered transmission rate exceeds the policing rate based on the packet information sent by the transmitting terminal;
Means for setting the value obtained as a new retransmission timeout value when the value obtained by subtracting the time obtained by the measuring means from the time when the bucket size is satisfied from 0 to the upper limit value is larger than the round-trip delay time. A communication device characterized by the above. The communication device according to claim 1,
The retransmission timeout value setting means includes
When a retransmission timeout occurs, the retransmission timeout value is temporarily changed using a binary search method with the current retransmission timeout value as an initial value to search for a retransmission timeout value that maximizes the throughput. A communication device, characterized in that it is means for setting a maximized retransmission timeout value as a new retransmission timeout value. The communication device according to claim 4, wherein
The search termination condition of the binary search method is that the throughput has not improved, the throughput has reached the value set as the target for the policing rate, or the retransmission timeout change width is smaller than a predetermined threshold. A communication apparatus, characterized by The communication device according to claim 1,
The change of the retransmission timeout value is performed at an arbitrary predetermined interval when a retransmission timeout occurs in a range from a round-trip delay time to a generally set retransmission timeout value. Communication device.   A program that causes a computer to function as each unit in the communication device according to any one of claims 1 to 6. When TCP communication is performed, the same transmission / reception address information and port information are managed as one flow in the transmission side environment, and the retransmission timeout is detected for each flow and the throughput until the retransmission timeout occurs is measured. Throughput measurement recording procedure to record,
A retransmission timeout value setting procedure for changing a retransmission timeout value, obtaining a retransmission timeout value that maximizes the throughput by measuring the throughput by the throughput measurement recording procedure, and setting it as a new retransmission timeout value;
A communication procedure for performing communication based on the set maximized retransmission timeout value. The communication method according to claim 8, wherein
To obtain the policing settings that are set for the network to which the token bucket policer is applied,
A measurement procedure for measuring the time from when the first retransmission timeout occurs until the recovered transmission rate exceeds the policing rate based on the packet information sent by the transmitting terminal,
And a procedure of setting the value as a new retransmission timeout value when the value obtained by subtracting the time obtained by the measuring means from the time when the bucket size is satisfied from 0 to the upper limit value is larger than the round-trip delay time. A communication method characterized by the above. The communication method according to claim 8, wherein
The retransmission timeout value setting procedure includes:
When a retransmission timeout occurs, the retransmission timeout value is temporarily changed using a binary search method with the current retransmission timeout value as an initial value to search for a retransmission timeout value that maximizes the throughput. A communication method characterized by a procedure for setting a maximized retransmission timeout value as a new retransmission timeout value.

Images