JP2015050746A - Packet communication device and system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a packet transmitter which can distribute packets to a suitable link from one or more viewpoints.SOLUTION: A packet transmitter transmits a packet to one of a plurality of links and can correspond to packet retransmission when packet loss occurs in each link. The packet transmitter includes: a switch function section for distributing the packet to one of the links; and a transfer optimization function section which acquires at least one of first information which shows how calculation resources in the transmitter are distributed to transfer processing to the individual links, second information which shows characteristics and states of the individual links and third information which shows a content that a transmission source of the packet requests of transfer, and determines a distribution method of the packet by the switch function section to indicate it in response to the acquired information.

Description

  The present invention relates to a packet communication apparatus and system, and can be applied to, for example, a case where communication of one TCP (Transmission Control Protocol) session is performed using a plurality of paths and interfaces.

  There is a known communication technique defined in the standard RFC6824 by the standard organization IETF (see Non-Patent Document 1). This technology communicates one TCP session using a plurality of links when performing communication by TCP between terminals or between terminals and servers using the Internet, and is generally called MPTCP. When MPTCP is used, communication can be performed at a higher speed than conventional TCP, and reliability in communication using an unstable transmission medium such as wireless communication can be improved.

  FIG. 4 shows a configuration of a packet communication system (hereinafter referred to as an MPTCP application system) 3 to which MPTCP is applied. FIG. 4 shows a one-way packet communication system for ease of explanation. FIG. 4 shows a case where the number of links is five, and the expression “five” in the following description indicates the number of links.

  The MPTCP-compatible packet transmission device 1 includes an outer transmission TCP function unit (denoted as TCP (outer) in FIG. 4) 10, a switch function unit (denoted as SW in FIG. 4) 11, and five link-specific queues (in FIG. 4). Q) 12-1 to 12-5, five inner transmission TCP function units (denoted as TCP in FIG. 4) 13-1 to 13-5, five transmission interface units (denoted as IF in FIG. 4) 14 -1 to 14-5.

  On the other hand, the MPTCP-compatible packet receiver 2 includes five reception interface units (indicated as IF in FIG. 4) 20-1 to 20-5, and five queue-specific queues (indicated as Q in FIG. 4) 21-1 to 21. -5, five inner reception TCP function units (indicated as TCP in FIG. 4) 22-1 to 22-5, integrated queue (indicated as Q in FIG. 4) 23, outer reception TCP function unit (in FIG. 4, TCP (Denoted as (outside)) 24.

  In a packet communication system (hereinafter referred to as a TCP application system) to which a normal TCP is applied, the MPTCP application system 3 is different from the TCP application system in that there is one TCP function unit on each of the transmission side and the reception side. In addition to the same TCP function units 10 and 24, a plurality of TCP function units 13-1 and 22-1,..., 13-5 and 22-5 that are paired on both the transmission side and the reception side Exists.

  In the packet transmission device 1, a packet transmitted from application software (hereinafter referred to as application software) 15 is received by the outer transmission TCP function unit 10 and transmitted toward the reception side. The application software 15 and 25 are not constituent elements of the packet transmission device 1 and the packet reception device 2, but are written for explanation.

  The outer transmission TCP function unit 10 divides the information received from the application software 15 into an MTU (Maximum Transmission Unit) size, and transmits it with an order number. Data is transmitted up to the full size of the smaller of the transmit and receive windows. The reception window size is the remaining capacity of the integrated queue 23 immediately before the outer reception TCP function unit 24, and if it is notified that this is 0, the outer transmission TCP function unit 10 on the transmission side stops transmission. . Further, the outer reception TCP function unit 24 assembles the received packets in the order of the order numbers, and sends order number information as a delivery confirmation signal (hereinafter referred to as an Ack signal) to the transmission side for delivery confirmation. Within the network, packet order reversal and packet dropping sometimes occur. When the sequence number is different from the expected value, the outer reception TCP function unit 24 returns the Ack signal with the previous sequence number of the packet that has arrived validly to the transmission side. In the outer transmission TCP function unit 10, when the Ack signal into which the same sequence number is inserted arrives continuously and the number of consecutive times exceeds the threshold, it is considered to have been discarded and retransmitted. The functions and operations described above are the same as those of the TCP application system.

  In the MPTCP application system 3, the switch function unit 11 is present at the subsequent stage of the outer transmission TCP function unit 10, and the switch function unit 11 includes the queues 12-1 to 12-5 and the inner transmission TCP function units 13-1 to 13. The input packet is selectively transferred to any of the five inner transmission function pairs consisting of -5. The inner transmission TCP function units 13-1 to 13-5 are respectively connected to the inner reception TCP function of the opposing packet reception device 2 via the interface units 14-1 to 14-5 and 20-1 to 20-5. It communicates with the sections 22-1 to 22-5. Since the MPTCP application system 3 uses five links at the same time, the maximum window size is the same, the number of queues 12-1 to 12-5 is five times, and when the speed on the application software 15 side is sufficiently high, More information can be transferred than transmitted over a link. Window control, delivery confirmation, and retransmission control are also performed between the inner transmission TCP function units 13-1 to 13-5 and the inner reception TCP function units 22-1 to 22-5 located on both sides of these five links. .

  Hereinafter, each pair of TCP session portions between the inner transmission TCP function units 13-1 to 13-5 and the inner reception TCP function units 22-1 to 22-5 may be referred to as an inner TCP session. The flow there is sometimes called a subflow. In addition, a TCP session between the outer transmission TCP function unit 10 and the outer reception TCP function unit 24 may be referred to as an outer TCP session.

  The number on the left side of the link in FIG. 4 indicates the sequence number of the packet sent out by the packet transmitting apparatus 1, and the number on the right side of the link in FIG. 4 indicates the order number of the packet that reaches the packet receiving apparatus 2. The packet with the sequence number “2” is discarded during transmission and retransmitted. In addition, in FIG. 4, the propagation delay of the second link from the bottom is larger than that of the bottom link, so that even if it is sent on the bottom link later, it arrives earlier than it is sent on the second link from the bottom. In view of the above, the switch function unit 11 selects the inner transmission function pair so as to correspond to the first, second, third, fifth, and fourth links from the top, and the order of packets that arrive at the receiving side is The case where it is made to become the packet from the 1st, 2nd, 3rd, 4th, 5th link is shown.

  The MPTCP application system 3 has a large differentiation point that is left to implementation. The switch function unit 11 is connected to which inner transmission TCP function unit 13-1 to 13-5 (in other words, the queues 12-1 to 12-5 and the inner transmission TCP function units 13-1 to 13- It is the best method (hereinafter referred to as a link selection method) to give a packet to the inner transmission function pair (5) and communicate.

  In this regard, Non-Patent Document 2 describes the following four link selection methods.

  The first is a round robin method in which every link is used one by one so as to be used equally.

  The second is a method in which priority is given to reducing arrival order reversal and the RTT (Round Trip Time) is sent to the link having the minimum round trip time.

  The third is to add a value obtained by dividing the amount of traffic accumulated in the transmission queue by the throughput to the RTT, and make a selection based on the sum of the transfer delay and the expected time to leave the queue to further reduce the possibility of arrival order reversal. This is an aimed ATLB (Arrival-Time Matching Load-Balancing) method.

  Fourth, in the third ATLB method, in order to improve the accuracy of the prediction of the time to exit the queue, not only the traffic amount is divided by the throughput but also the phase of TCP is reflected.

RFC6824: TCP Extensions for Multiple Operation with Multiple Addresses Oikawa Nagatoshi et al., “Proposal of Multipath TCP Packet Distribution Method in Network with Large Delay Difference”, IEICE Technical Report NS2012-106, November 2012

  However, the four link selection methods described above have the following problems.

  The first problem is that CPU resources (processing capability) and memory resources (storage capability) for each of the inner transmission TCP function units 13-1 to 13-5 and each of the inner reception TCP function units 22-1 to 22-5 This is a problem caused by inappropriate distribution.

  In the second to fourth link selection methods, one or a plurality of CPUs simultaneously process a plurality of TCP processes. If the CPU and memory resources applied to each TCP process are completely equal, the conventional link selection method is expected to work effectively. However, this is not always true. For example, if it is a specification to distribute a large CPU resource and memory resource (sometimes collectively referred to as a computer resource) to a process with a larger load, the inner TCP session where the packet arrived earlier will be The internal TCP session that has large CPU resources and memory resources, and the inner TCP session that arrives later can be allocated with less resources, and the inner TCP session that has received the packet later has less processing power, so it can process fewer packets than others. become. On the other hand, the inner TCP session given more resources can be processed more, more traffic flows, and as a result, feedback is given to allocate more resources, and the difference spreads more and more. It will be.

  This occurs on both the transmitting and receiving sides. In other words, a subflow with a small computer resource that can be used to transmit a packet on the transmitting side may have a smaller CPU time for transmission processing than a subflow with a large computer resource, and may reduce the number of packets that can be transmitted. Also on the receiving side, a subflow with a small CPU time that can be used for a receiving process or a process for returning an Ack signal to the transmitting side may reduce the number of packets that can be transmitted with a subflow with a large CPU time.

  The second problem is a problem that occurs when there is a power saving link.

  In recent years, a technology has been developed that reduces power consumption by changing the usable bandwidth of a physical link in accordance with the amount of traffic and reducing the bandwidth when the traffic is low. However, there are some implementations in which the TCP function unit cannot recognize this for both transmission and reception, and does not increase the transmission rate when the remaining bandwidth becomes small. As a result, the physical link is operated below the threshold for increasing the bandwidth, the physical link is operated with the bandwidth kept small without knowing that the user really wants to send more, and the TCP function side can actually send more However, if there is a possibility that congestion will occur when sending more than this, there is a possibility of operation without increasing the transmission rate. As a result, there is a possibility that the amount of packets sent from the MPTCP application system to the inner TCP session including the power saving link as the subflow link is smaller than the optimum value.

  A third problem is a problem that occurs when a difference in propagation delay time of a link related to the inner TCP session (hereinafter, sometimes abbreviated as a delay difference) is large.

  In the second to fourth link selection methods, if the delay difference is large, a link having a large propagation delay time is not used even if there is a surplus band, and there is a problem that throughput does not increase. For example, when there is a link LA that has little delay but a large delay time to arrival even when a large amount of traffic is passed, and a link LB that has a small delay time but tends to be discarded when the input traffic becomes large to some extent, In some cases, it is better to actively use the link LA according to the requirements of the application software and the queue length that can be secured. The above second to fourth link selection methods cannot cope with such a case.

  The fourth problem is related to the response to resource selection according to the required quality.

  It is assumed that there are two types of links, a link LC and a link LD, between the packet transmitter and the packet receiver, and the link LC has a higher usage cost than the link LD. Here, there are several types of application software to be used, the application software AP1 has a request to send even as fast as 0.1 seconds, and the application software AP2 has sufficient requirements at a speed of several bits / sec. Suppose that the application software AP3 only needs to be delayed within a certain period of time.

  It seems that the application software AP1 uses both the link LC and the link LD to control the traffic to the link LC and the link LD so that transmission can be performed in a minimum time. The application software AP2 seems to be preferable to minimize the cost by using only the link LD. It is considered that the application software AP3 preferably uses the link LD as much as possible within the range satisfying the required conditions, and uses the link LC within the range unavoidable to satisfy the required conditions, and aims to minimize the use cost.

  That is, even when there are a plurality of types of application software used for packet communication and the required qualities are different, only a certain link selection method can be applied, and this method also results in unsatisfactory application software.

  The present invention has been made in consideration of the above points, and is intended to provide a packet communication apparatus and system capable of distributing and transferring packets to a suitable link from one or more viewpoints. .

  According to a first aspect of the present invention, there is provided the packet transmission device in a packet communication system capable of handling packet retransmission at the time of packet loss, wherein a plurality of links for packet transfer are interposed between the packet transmission device and the packet reception device. In the corresponding packet communication device, (1) a switch function unit that distributes a packet to be transmitted to any one of the links, and (2) at least which computing resource in the packet transmission device is used for transfer processing to each link. 3 of the 1st information showing how it is distributed, the 2nd information showing the characteristic of each above-mentioned link, the 2nd information showing the state, and the 3rd information showing the contents which the sending origin of the packet which is going to transmit demands for transfer At least one of the two pieces of information is acquired, and based on the acquired information, the switch function unit distributes the input packet. That method to determine and having a transfer optimization unit that instructs the switching function portion.

  According to a second aspect of the present invention, there is provided a packet communication apparatus in which a plurality of links for packet transfer are interposed between a packet transmission apparatus and a packet reception apparatus, and each link can handle packet retransmission at the time of packet loss. As described above, the packet communication device according to the first aspect of the present invention is applied.

  According to the present invention, it is possible to provide a packet communication apparatus and system that can distribute and transfer packets to a link that is suitable from one or more viewpoints.

It is a block diagram which shows an example of the positional relationship on the network of the packet transmitter which comprises the MPTCP application system of embodiment, and a packet receiver. It is a block diagram which shows the structure of the MPTCP application system of embodiment. It is a flowchart which shows the operation | movement which the MPTCP optimization function part in the MPTCP application system of embodiment performs. It is a block diagram which shows the structure of the conventional MPTCP application system.

(A) Main Embodiment An embodiment of a packet communication apparatus and system according to the present invention will be described below with reference to the drawings.

  The packet communication system of the embodiment is an MPTCP application system. Most of the packet communication devices constituting the packet communication system have both a data packet transmission configuration and a reception configuration. However, in the following description of the embodiments, a pair of packet transmission devices is provided for the sake of simplicity. The packet receiver will be described.

(A-1) Configuration of Embodiment FIG. 1 is a block diagram illustrating an example of a positional relationship on the network between the packet transmission device 1A and the packet reception device 2A that configure the MPTCP application system 3A of the embodiment.

  In FIG. 1, the packet transmitting apparatus 1A and the packet receiving apparatus 2A can be connected via a virtual network 5 having an SDN (Software Defined Networking) node 4 (hereinafter simply referred to as an SDN node) using an open flow. Has been made. In FIG. 1, links (routes) LK1 to LK3 of three inner TCP sessions are shown, and each link LK1 to LK3 passes through one or a plurality of SDN nodes 4, respectively. The link LK1 is routed through the SDN nodes 4-1, 4-2 and 4-3, the link LK2 is routed through the SDN nodes 4-1 and 4-3, and the link LK3 is routed to the SDN node. 4-1, 4-4, and 4-3.

  A transmission-side information processing device 6A is provided in association with the packet transmission device 1A. For example, the packet transmission device 1A may be mounted on the transmission-side information processing device 6A, or may be constructed as a separate device from the transmission-side information processing device 6A. For example, the same personal computer may serve as both the transmission-side information processing device 6A and the MPTCP-compatible packet transmission device 1A, and the following description will be given assuming such a case.

  Similarly, a receiving side information processing device 7A is provided in association with the packet receiving device 2A. In the following description, it is assumed that the same personal computer serves as both the receiving side information processing apparatus 7A and the MPTCP compatible packet receiving apparatus 2A.

  FIG. 2 is a block diagram illustrating the configuration of the MPTCP application system 3A according to the embodiment, in which the same and corresponding parts as those in FIG. Unlike FIG. 1, FIG. 2 shows five links as links between the packet transmission device 1A and the packet reception device 2A.

  As described above, when both the packet transmitting device 1A and the packet receiving device 2A are constructed on a personal computer, most of the components are often composed of a CPU and software executed by the CPU. Even in such a case, it can be functionally represented in FIG.

  In FIG. 2, the packet transmission device 1A according to the embodiment includes an outer transmission TCP function unit 10, a switch function unit 11, five link-specific queues 12-1 to 12-5, and five inner transmission TCP function units 13-1. 13-5, in addition to the five transmission interface units 14-1 to 14-5, the transmission side MPTCP optimization function unit 16 and the transmission side computational resource management function unit 17 are provided.

  The packet reception device 2A according to the embodiment includes five reception interface units 20-1 to 20-5, five link-specific queues 21-1 to 21-5, and five inner reception TCP function units 22-1 to 22-5. In addition to the integrated queue 23 and the outer receiving TCP function unit 24, the receiving side MPTCP optimization function unit 26 and the receiving side computing resource management function unit 27 are provided.

  Here, the outer transmission TCP function unit 10, the switch function unit 11, the link-specific queues 12-1 to 12-5, the inner transmission TCP function units 13-1 to 13-5, and the transmission interface units 14-1 to 14- 5, reception interface units 20-1 to 20-5, link-specific queues 21-1 to 21-5, inner reception TCP function units 22-1 to 22-5, integrated queue 23, and outer reception TCP function unit 24 The basic functions are the same as those in FIG. 4 and will be briefly described below.

  The outer transmission TCP function unit 10 performs packet transmission / retransmission and transmission window control for the outer TCP session, and the outer reception TCP function unit 24 performs packet reception, delivery confirmation, and retransmission control for the outer TCP session. , Order adjustment, and reception window control. The switch function unit 11 distributes a packet to be transmitted to one of the inner TCP sessions (in other words, a link). The inner transmission TCP function unit 13-n (n is 1 to 5) performs packet transmission / retransmission and transmission window control for the inner TCP session related to the link LKn. The inner reception TCP function unit 22-n Performs packet reception, delivery confirmation, retransmission control, order adjustment, and reception window control for the inner TCP session related to the link LKn. The link-specific queue 12-n queues packets waiting to be transmitted by the inner transmission TCP function unit 13-n. The transmission interface unit 14-n performs a transmission-side interface for the link LKn, and the reception interface unit 20-n performs a reception-side interface for the link LKn. The link-specific queue 21-n queues packets waiting to be processed by the inner reception TCP function unit 22-n. The integrated queue 23 accepts packets related to the outer TCP session from the inner reception TCP function units 22-1 to 22-5 and queues packets waiting for processing by the outer reception TCP function unit 24. is there.

  The transmission-side computational resource management function unit 17 and the reception-side computational resource management function unit 27 that are newly provided in the embodiment are the usage statuses of computational resources such as the CPU, memory, and bus used by the packet transmission device 1A and the packet reception device 2A, respectively. It is something that manages. Each of the computing resource management function units 17 and 27 may be newly constructed for the devices 1A and 2A, but if the devices 1A and 2A are constructed on an information processing device such as a personal computer. The computing resource management function unit included in the OS (Operating System) of the information processing apparatus may be applied as it is.

  The transmission side MPTCP optimization function unit 16 optimizes the allocation of packets to the links (and therefore the inner TCP sessions) LK1 to LK5. The reception side MPTCP optimization function unit 26 includes a transmission side MPTCP optimization function. The receiving side cooperates with the optimal allocation in the unit 16. The transmission side MPTCP optimization function unit 16 basically instructs the switch function unit 11 on how to determine the inner TCP session, but in order to collect information necessary for issuing the determination method instruction. An instruction is given to the outer transmission TCP function unit 10, the inner transmission TCP function units 13-1 to 13-5, and the computing resource management function unit 17. Similarly, the receiving-side MPTCP optimization function unit 26 collects necessary information, so that the outer receiving TCP function unit 24, the inner receiving TCP function units 22-1 to 22-5, and the computational resource management function unit 27 are collected. Or give instructions.

  The transmission side MPTCP optimization function unit 16 includes a communication function unit 16a, a calculation resource allocation grasping function unit 16b, a link characteristic grasping function unit 16c, and an optimum switching instruction function unit 16d, and the receiving side MPTCP optimization function unit 26 includes: A communication function unit 26a, a calculation resource allocation grasping function unit 26b, and a link characteristic grasping function unit 26c are included.

  The communication function unit 16a is connected to the outer transmission TCP function unit 10, and the communication function unit 26a is connected to the outer reception TCP function unit 24, and transfers control information (control packet) between the communication function units 16a and 26a. It's a good thing. The communication function unit 26a is mainly a control information transmission side, and the communication function unit 16a is a control information reception side.

  The computing resource allocation grasping function unit 16b on the transmission side accesses the computing resource management function unit 17, and the computing resources such as the CPU, memory, bus, etc. grasped and operated by the computing resource management function unit 17 are connected to each inner TCP. Information on how much the session (and hence the link) is allocated is obtained and given to the optimum switching instruction function unit 16d. Similarly, the computing resource allocation grasping function unit 26b on the receiving side accesses the computing resource management function unit 27 to acquire information on the allocation state of computing resources, and receives the communication function units 26a and 16a on the receiving side and the transmitting side. Further, it is given to the optimum switching instruction function unit 16d via the calculation resource allocation grasping function unit 16b on the transmission side. As described in the section of the problem of the conventional MPTCP application system 3, the performance of the TCP subflow depends on the calculation resources on both the transmission side and the reception side. Therefore, it is necessary to measure with both computational resources.

  The link characteristics grasping function units 16c and 26c on the transmission side and the reception side cooperate to provide physical bandwidth of each link LK1 to LK5, presence / absence of power saving control (control to reduce the speed when there is little traffic), power saving control The usable bandwidth including the packet, the propagation delay time, and the packet discard rate are measured (some items may be predicted values). In addition, the link characteristic grasping function units 16c and 26c are the sizes of the transmission-side and reception-side queues 12-1 to 12-5 and 21-1 to 21-5 on the inner TCP session, and the window size. Obtain phase information of the inner TCP session. The link characteristic grasping function units 16c and 26c give the acquired information to the optimum switching instruction function unit 16d. Note that control information and measurement result information used for measurement are appropriately transferred via the communication function units 16a and 26a.

  The optimum switching instruction function unit 16d determines an inner TCP session (subflow) to which the switch function unit 11 transfers a packet, based on the allocation state of the calculation resources and the link characteristics, and determines each inner TCP session determined. A method of sorting is determined and an instruction is given to the switch function unit 11. The switch function unit 11 of this embodiment gives a packet to the inner transmission TCP function units 13-1 to 13-5 of an appropriate inner TCP session (subflow) according to the instructed method (note that Queued accordingly).

(A-2) Operation of Embodiment Next, the operation of the MPTCP application system 3A of the embodiment having the above configuration will be described. The MPTCP application system 3A according to the embodiment is significantly different from the conventional system in that operations performed in cooperation with the MPTCP optimization function units 16 and 26 are added. The operations performed in cooperation with each other will be described.

  FIG. 3 is a flowchart showing an operation performed by the MPTCP optimization function unit 50 (reference numeral 50 is 16 and / or 26). The operations shown in FIG. 3 (steps S100, S200, and S300) are repeatedly executed at predetermined time intervals, for example (the predetermined time may be variable as described later).

  The MPTCP optimization function unit 50 acquires information related to communication (MPTCP communication) of a plurality of inner TCP sessions (subflows) currently or in the future (step S100), and based on the acquired information, the switch function Quality is estimated for each switching method candidate by the unit 11 to determine an optimal switching method (step S200), and control is performed so that the switch function unit 11 operates according to the determined optimal switching method (step S300). I do.

  The operation for acquiring information necessary for MPTCP communication includes an operation for grasping the allocation status of computing resources (step S101), an operation for grasping characteristics of a link corresponding to each inner TCP session (step S102), and application software. An operation (step S103) of obtaining the requested quality information.

  In step S101, the computing resource allocation grasping function units 16b and 26b access the computing resource management function units 17 and 27, respectively, and the computing resources managed and operated by the computing resource management function units 17 and 27 are Information about how much is allocated to the inner TCP session (and therefore the link) is obtained and provided to the optimum switching instruction function unit 16d.

  In step S102, the link characteristics grasping function units 16c and 26c on the transmission side and the reception side cooperate to link the physical band of each link LK1 to LK5, presence / absence of power saving control, usable band including power saving control, propagation In addition to measuring the delay time and the packet discard rate, the link characteristic grasping function units 16c and 26c are respectively linked to the link-specific queues 12-1 to 12-5 and 21-1 to 21-5 related to the inner TCP session. The size and the phase information of the inner TCP session are acquired and provided to the optimum switching instruction function unit 16d.

  In step S103, the optimum switching instruction function unit 16d acquires the quality information requested by the application software 15 directly from the application software 15 or via the outer transmission TCP function unit 10. For example, the quality information required for the application software 15 is described, and the optimum switching instruction function unit 16d captures the description content. The application software 15 describes quality information (requirement conditions) using, for example, the link speed and billing (fee system) as parameters. (Example 1) If the delay time is as short as possible, the pay-per-use link is preferentially used even if it is charged according to the traffic. (Example 2) The flat-rate charging link is preferentially applied within the range satisfying the service quality. The quality is requested by using a pay-per-use link only when it is unavoidable to satisfy the quality, (Example 3) using only a flat-rate link regardless of the delay time and the packet loss rate, etc. Information for specifying a distribution policy, which will be described later, may be included in the quality information requested by the application software 15. Alternatively, by limiting the distribution policy that can be applied to one or more types depending on the link type, when the type of link to be applied is determined as quality information, one or more types of distribution policies that can be applied are determined. (For example, the optimum switching instruction function unit 16d incorporates information for obtaining a distribution policy from the required quality information).

  Based on the acquired information, the operation of estimating the quality for each switching method candidate by the switch function unit 11 and determining the optimum switching method includes the quality estimation operation for each switching method candidate (step S201) and the optimal switching. This is divided into an operation for determining a method (step S202). In the above explanation, the link usage fee was used as the cost condition when selecting the link. However, another policy such as “(use rival company A's link as much as possible)” is used for evaluation. You may use as cost.

  In step S201, the optimum switching instruction function unit 16d uses N (N is 1 to M) of all M links (5 in the case of FIG. 2) based on the information acquired in step S100. Thus, the computing resources necessary for communication of each link are distributed, and the communication quality for each link when the packet is distributed to each link and transferred by each candidate distribution policy is estimated. The calculation resource is distributed according to, for example, a transmission band for each link (or an available band for each link).

  In the process of step S201, the communication quality may be estimated for all combinations of selecting M to N (N is 1 to M).

  However, the following other N determination methods may be applied. For example, the N links may be all links applicable at that time regardless of the request of the application software 15. Further, the N links may be narrowed down according to the quality information requested by the application software 15 (for example, if the request uses only a flat-rate charge link, the N links are fixed-rate charge). Only links).

  The communication quality for each link is, for example, a used band, a transfer time, another parameter, or a weighted average of a plurality of parameters. Further, the communication quality is calculated as an average value when it is assumed that a predetermined number of packets are distributed, for example.

  As a distribution policy, it is assumed that a plurality of candidates as exemplified below are prepared, but there may be only one distribution policy.

  The first candidate distribution policy (hereinafter referred to as distribution policy A) is a policy in which packets are distributed to a link having a minimum transfer time (or less than a predetermined threshold) expected at the time of distribution. For example, if the switch function unit 11 does not have a transfer time calculation function, each time a packet is distributed, the link information of the minimum transfer time can be obtained from the optimum switching instruction function unit 16d. good.

  The distribution policy of the second candidate (hereinafter referred to as distribution policy B) is assigned to each link so that the input interval to each link is constant according to the size of the available bandwidth estimated to be transferable on each link. The policy is to distribute packets. For example, in the case where the links LK1 to LK5 are used and the ratio of the available bandwidth is approximately 2: 1: 1: 1: 1, LK1- By allocating as LK2-LK3-LK1-LK4-LK5, the input interval is also made constant for the link LK1. Note that, as described above, the distribution policy B includes a link in which packet transfer is larger than other links, and there is a risk of erroneous determination of packet loss due to order reversal. To avoid such misjudgment, if distribution policy B is selected, relax the criteria for determining packet loss due to arrival of the same number of Ack signals (increase the number of arrivals determined to be packet loss). Is preferred. In addition, the distribution policy B has a concern that the detection timing of the change in the link state is later than the distribution policy A. For example, when the distribution policy B is selected, the waiting time until the process of step S300 ends and the next step S100 is started is shorter than when the distribution policy A or a distribution policy C described later is selected. You may make it do.

  The third candidate distribution policy (hereinafter referred to as distribution policy C) is a policy of distributing packets in accordance with a simple round robin method.

  The estimated transfer time of the distribution policy A and the delay time calculated as the communication quality for each link (hereinafter referred to as the link-specific transfer time) are the propagation delay time of the link and the link-specific queues on the transmission side and the reception side. This is the total time of the queue delay time required to pass through and the TCP processing delay time of the inner TCP session. In the case of this embodiment, when calculating the link-specific transfer time, the following processing is performed.

  When calculating the TCP processing delay time, the optimum switching instruction function unit 16d uses the delay time including the TCP phase information. Further, when there is a power saving link, it is assumed that there is no power saving control, and the transfer time for each link is obtained on the condition that calculation resources are allocated to each link according to a predetermined rule. For example, calculation is performed by prolonging the TCP processing delay time that does not consider the allocation amount of each link of calculation resources by a ratio (for example, extracted by applying a conversion table) according to the allocation amount of each link of calculation resources. You may make it reflect the allocation amount of each link of a resource. When there is a power saving link, the optimum switching instruction function unit 16d determines that the available bandwidth of the link is not an available bandwidth that can be used at the present time, but from the capacity of the link that can be used without a power saving operation. The bandwidth is excluded from the bandwidth used in. Since the calculation of the queue delay time is affected by the bandwidth per unit time that can be sent to the link, the change in the method for determining the available bandwidth of the power saving link has an effect.

  For example, the available bandwidth of the link used by the distribution policy B and the bandwidth calculated as the communication quality for each link (hereinafter referred to as the link-specific bandwidth) are calculated as follows, for example. In the following, a calculation example in the case of TCP-reno is shown, but in the case of other TCP versions, a calculation formula corresponding to the version may be applied.

The communication speed (bandwidth) at a short time is obtained by equation (1), and the communication speed at times Ta to Tb is obtained by equation (2).

  Here, when the TCP session is in the slow start phase, the window size Ws (m) at the time Tk applied to the expression (2) is the initial window size W0, and the number of times of receiving the Ack signal between T0 and Tk. When m, it follows Formula (3). In the equation (3), 2 ^ m represents 2 to the power of m.

Ws (m) = W0 × 2 ^ m (3)
When the TCP session is in the congestion avoidance phase, the window size Wc (n) at the time Tk applied to the expression (2) is the window size immediately after entering the congestion avoidance phase as Wc0, and after entering the congestion avoidance phase. When the number of times of receiving the Ack signal is n and the packet is not discarded, the equation (4) is followed.

Wc (n) = Wc0 + W0 × n (4)
In the case of the slow start phase, when packet discard occurs or when a duplicate Ack signal is detected, the process proceeds to the congestion avoidance phase. In the congestion avoidance phase, if the packet number inserted in the Ack signal when the duplicate Ack signal due to discard is detected is p, the window size Wc (p + 1) of the link when sending the p + 1th packet is It is represented by the formula (5).

Wc (p + 1) = Wc (p) × (1/2) (5)
The optimum switching instruction function unit 16d may allocate communication resources according to the available bandwidth and calculate the communication quality. Further, if there is a margin in the usable bandwidth, the window size may be enlarged within the range to process. When there is a power saving link, the window size may be enlarged to the limit of the available bandwidth.

  In step S202, the optimum switching instruction function unit 16d obtains a combination of a link and a distribution policy that satisfies the quality information required by the application software 15, and among them, the cost, which is an evaluation value indicating that the smaller one is better, is the smallest. Is determined as a combination of a link to be applied and a distribution policy.

  Here, the cost is the sum of the costs for the N links. The cost of each link is a weighted addition value between the link usage fee and the link communication quality obtained in step S201. The weight defines not only the function of aligning the dimensions of the usage fee and the communication quality, but also whether the weight is placed on the usage fee or the communication quality. Here, the weight for the usage fee and the weight for the communication quality may be fixed, but they are switched according to the required quality information of the application software 15 or the type of the application software 15 (telephone communication, content distribution, etc.). Is preferred. For example, the weight is switched by applying a conversion table that can extract the weight using the required quality information or the type of application software 15 as a key. In the case of the application software 15 that places importance on the usage fee, the weight for the usage fee is made smaller than the weight for the communication quality so that the usage fee contributes to low cost (high evaluation), and the communication quality (for example, communication speed) is reduced. In the case of the application software 15 to be emphasized, the weight for the communication quality is made smaller than the weight for the usage fee so that the communication quality contributes to low cost (high evaluation).

  In step S300, the optimum switching instruction function unit 16d instructs the switch function unit 11 to transfer the packet with the distribution policy determined for the N links determined in step S202. As a result, the switch function unit 11 transfers the packet to the instructed N links (directly queues for each link) according to the instructed distribution policy.

(A-3) Effect of Embodiment According to the above-described embodiment, an MPTCP application system that solves the conventional problems can be realized.

  First, since information on how much computing resources are allocated to each inner TCP session is obtained and reflected in link selection, link selection becomes inappropriate due to inappropriate distribution of computing resources. This can be avoided.

  Secondly, the presence or absence of the power saving link is confirmed, and for the power saving link, the available bandwidth is adjusted and reflected in the selection of the link, so that the packet can be appropriately transferred to the power saving link, The entire link can be applied in a balanced manner.

  Third, link communication quality is not evaluated simply by link propagation delay time, but is also evaluated taking into account available bandwidth, queue waiting time, TCP processing time, etc. Just because the delay time is large, the link application rate does not decrease, and the entire link can be applied in a balanced manner.

  Fourth, since the required quality of the application software is grasped and reflected in the selection of the link, the link can be selected so as to satisfy the required quality of the application software.

(B) Other Embodiments In the description of the above-described embodiment, various modified embodiments have been referred to. However, modified embodiments as exemplified below can be cited.

  In the above-described embodiment, the calculation resource is distributed to the inner TCP session (process) of each link according to the transmission bandwidth to the link (or the usable bandwidth of the link), and reflected in the selection of the link. However, the computing resource may be fixedly distributed to (in the process of) the inner TCP session related to each link. For example, in the case where the OS of the information processing apparatus equipped with the packet transmission device 1A or the packet reception device 2A performs the power saving control, the power saving control is performed according to the fixed distribution of the calculation resources as described above. The inconvenience of distribution associated with can be avoided.

  In the above-described embodiment, there are three types of distribution policies. However, it is also possible to construct an MPTCP application system so that only one distribution policy is applied and the distribution policy cannot be selected. For example, when the distribution policy A is adopted as the only distribution policy, the above-described inconvenience (third problem) due to the propagation delay time difference can be avoided. Further, for example, when the distribution policy B is adopted as the only distribution policy, the above-described inconvenience (second problem) due to the existence of the power saving link can be avoided.

  In addition, a distribution policy other than the distribution policy described above may be introduced. For example, a distribution policy that determines the distribution of packets on each link may be introduced after ensuring that the computational resources are distributed in an amount greater than that required for the transfer process to each link. For example, in the case of using together with the distribution policy B, when the ratio of the number of packets to each link is determined from the available bandwidth of each link, it is confirmed whether or not the above-mentioned condition for the calculation resource is satisfied, and the link that does not satisfy the use You may make it change to not doing.

  In the above-described embodiment, the general-purpose device corresponding to a plurality of types of application software is shown in which the packet transmission device and the packet reception device of the embodiment are mounted. However, the packet transmission device and the packet reception device of the embodiment are It may be mounted on a dedicated device. For example, the packet transmission device according to the embodiment may be mounted on the content distribution device. In such a case, there is no concept of application software as a packet supply source, and the required quality for MPTCP communication can be regarded as one type. For example, in such a case, the usage charge and communication quality weight in the link cost may be fixed. In other words, when the usage is limited to one type or the installed application software is one type, a fixed weight with a weight applied to only one of the usage fee or communication quality is applied. Anyway.

  In the above embodiment, the calculation resource state of the packet transmission device and the packet reception device is reflected in the link selection. However, only the calculation resource state of the packet transmission device may be reflected in the link selection. good.

  When control that allocates a large amount of computing resources or control of power-saving links is performed, until the bandwidth that stabilizes the control is expanded, the link selection intended for computing resources and power-saving links can be performed according to the above embodiment. The transmitted packet may be discarded. In order to avoid such fear, dummy packets are transferred to the inner TCP session of the subflow that uses the link to which the above control is applied, and a load is applied to the link so that the packet can be transferred sufficiently. After that, a normal traffic packet may be transmitted instead of the dummy packet. Alternatively, the dummy packet may be mixed with the normal traffic packet and transmitted, and after the packet can be sufficiently transferred, the mixing of the dummy packet may be stopped.

  In the above-described embodiment, the link itself used for packet transfer is also selected. However, a plurality of links used for packet transfer may be fixedly determined. Further, a plurality of links used for packet transfer may be fixedly determined depending on the type of application software.

  In step 201 of the above embodiment, the communication quality is obtained and the operation whose control target is to maximize the throughput is shown. However, instead of the communication quality, the control is performed to minimize the power consumption. You may do it. For example, a subflow link that consumes the least power may be used, and a link that consumes a large amount of power may not be used. Alternatively, it is only necessary to suppress the transmission of a packet to a link that increases power consumption.

  In the above embodiment, the case where the present invention is applied to the MPTCP application system has been shown, but the application target of the present invention is not limited to this. In short, the present invention can be applied to any system that transfers packets between a packet transmission device and a packet reception device via a plurality of links.

1A ... packet transmitting device, 2A ... packet receiving device, 3A ... MPTCP application system,
DESCRIPTION OF SYMBOLS 10 ... TCP function part for outer side transmission, 11 ... Switch function part, 12-1 to 12-5 ... Queue according to transmission side link, 13-1 to 13-5 ... TCP function part for inner side transmission, 14-1 to 14- DESCRIPTION OF SYMBOLS 5 ... Transmission interface part, 16 ... Transmission side MPTCP optimization function part, 16a ... Communication function part, 16b ... Computation resource allocation grasp function part, 16c ... Link characteristic grasp function part, 16d ... Optimal switching instruction | indication function part, 17 ... Transmission Computing resource management function side,
20-1 to 20-5: Reception interface section, 211-1 to 21-5: Reception side link-specific queue, 22-1 to 22-5 ... Inner reception TCP function section, 23 ... Integration queue, 24 ... Outer reception TCP function unit, 26... Receiving side MPTCP optimization function unit, 26 a... Communication function unit, 26 b... Calculation resource allocation grasping function unit, 26 c.

Claims (9)

In the packet communication apparatus corresponding to the packet transmission apparatus in the packet communication system capable of handling packet retransmission at the time of packet loss, a plurality of links for packet transfer are interposed between the packet transmission apparatus and the packet reception apparatus.
A switch function unit that distributes packets to be transmitted to any of the above links;
First information indicating at least how the computing resources in the packet transmission apparatus are allocated to the transfer processing to each link, second information indicating the characteristics and state of each link, and transmission of the packet to be transmitted Acquire at least one of the three pieces of information of the third information representing the content originally requested for transfer, and based on the acquired information, the switch function unit determines a method for distributing the input packet. And a transfer optimization function unit for instructing the switch function unit.   The packet communication apparatus according to claim 1, wherein the transfer optimization function unit also determines one or more links to be applied to packet transfer based on the acquired information.   Based on the acquired information, the transfer optimization function unit estimates the size of the available bandwidth that can be transferred on each link, and determines the available bandwidth of each link according to the estimated available bandwidth size. 3. The packet according to claim 1, wherein a packet distribution ratio is determined, and the switch function unit is instructed to distribute the packets so that an interval between packets input to each link is constant. Communication device.   The transfer optimization function unit acquires the first information, and distributes the packet to each link so that the calculation resources are distributed in an amount more than necessary for the transfer process to each link. The packet communication apparatus according to claim 1, wherein the method is instructed to the switch function unit.   The transfer optimization function unit estimates communication quality in packet transfer using each link based on at least one of the acquired first and second information, and based on the acquired third information. Understand whether the content of the packet source that is about to be transmitted is cost-oriented or quality-oriented, and estimate the cost for each link determined in advance based on the content of the requested request The packet communication apparatus according to claim 1 or 2, wherein a packet distribution method for each link is determined by changing a weight with respect to the communication quality.   6. The packet communication apparatus according to claim 5, wherein the estimated communication quality is a size of an available band that can be transferred by each link.   The transfer optimization function unit prepares a plurality of distribution policies as distribution policies that define how to distribute packets to each link, and tries to transmit to a distribution method candidate determined by each distribution policy. Applying an evaluation value calculated based on whether the content requested by the packet source for transfer is cost-oriented or quality-oriented, and determines the final distribution method of packets to each link. The packet communication device according to claim 1 or 2.   The transfer optimization function unit, as one of a plurality of distribution policies, based on the acquired first information, so that the calculation resources are distributed more than necessary for the transfer process to each link. The packet communication apparatus according to claim 7, further comprising a policy for distributing packets to each link. In a packet communication system capable of supporting packet retransmission at the time of packet loss, a plurality of links for packet transfer are interposed between the packet transmission device and the packet reception device.
A packet communication device, wherein the packet communication device according to claim 1 is applied as the packet transmission device.

Images