JP2010219677A - Network simulation device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a network simulation device suitable for evaluating traffic volume or the like in a network including a P2P relay type distribution system. <P>SOLUTION: In the network having a plurality of clusters to which a plurality of nodes are connected, the clusters are divided into branch clusters 10 for relaying and leaf clusters 20 connected to only one branch cluster, an inter-cluster link between the branch clusters and a link between the branch cluster and the leaf cluster are shared by communication between the nodes, and an inter-node link between the nodes (piers 21) in the leaf cluster occupies a band for each node. Each node has queues of the inter-node link and the inter-cluster link together, only the data transmitted by the node in the cluster are output on the basis of the queueing of the inter-node link and the inter-cluster link, and thus the delay time is calculated only for an up link. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a network simulation apparatus for evaluating traffic volume, throughput, and the like in a network distribution system.

  The network simulation device models the network topology and traffic information from the planned network topology and devices used before actually introducing the system, and roughly estimates the traffic volume and throughput. Yes, it is done by running various software developed for simulation on the network.

  For example, as shown in Non-Patent Document 1 and Non-Patent Document 2, OPNET (trade name) and Qualnet (trade name) are network-to-network connections and various protocols that operate on terminals and relay devices on the network. (Terminal protocols such as TCP, UDP, HTTP, relay node protocols such as OSPF, BGP, etc.) can be defined.

  On the other hand, Artisoc (trade name) and N3 Simulator, which are complex simulators, are simulation devices for the purpose of simulating autonomous decentralization of end terminals, as shown in Non-Patent Document 3 and Non-Patent Document 4, It is possible to flexibly define the protocol on the terminal.

http://www.johokobo.co.jp/opnet/index.html http://www4.kke.co.jp/qualnet/ http://www.kke.co.jp/product/catalogue/111_index_msg.html IEICE Technical Report CPSY2002-32, pp.55-60

  The above-mentioned OPNET and Qualnet are network simulation devices for testing the operation of Internet protocols such as TCP / IP and OSPF / BGP on the network. Detailed simulations such as packet transmission / reception at terminals and queuing processing at routers are performed. it can.

  However, there is a problem that it is not possible to define a protocol in which individual terminals operate in an autonomous distributed manner such as BitTorrent. Further, since the amount of calculation increases when the scale of terminals and routers to be handled is large, there is a problem that the number of routers can be handled only up to about 1,000 when using a standard server PC.

  The above-mentioned Artisoc and N3 Simulator are simulators suitable for simulating end terminals that operate in an autonomous and distributed manner, and are not limited to Internet protocols, but have their own protocol for end terminals and event occurrence timing on end terminals. Can be defined.

  However, since it is specialized in the operation of the terminal, there has been a problem that it is impossible to simulate the behavior of links and routers on the network.

  The present invention has been proposed in view of the above circumstances, and is a network simulation suitable for evaluating the amount of traffic and throughput between nodes (a general term for terminals, routers, servers, etc.) in a network constructed with a P2P relay type distribution system. The object is to provide a device.

  In order to achieve the above object, according to the present invention, when defining a network, the roles of the relay network and the access network are separated, and only for the access network that is important in the simulation of the P2P relay type distribution system, By providing a function that faithfully reproduces the behavior, it is possible to simulate both the network and the terminal on a large scale with a small amount of calculation.

That is, the network simulation apparatus according to claim 1 is connected to only one branch cluster and one branch cluster for the plurality of clusters in a network having a plurality of clusters (small networks) to which a plurality of nodes are respectively connected. And the inter-node link between the branch clusters and between the branch cluster and the leaf cluster is shared by communication between nodes, and the inter-node link between the nodes in the leaf cluster is a node. Every band occupies a bandwidth.
Each node has both an inter-node link and an inter-cluster link queue, and outputs only the data transmitted by the nodes in the cluster in consideration of the waiting in the inter-node link and inter-cluster link queues. Thus, it is characterized in that the traffic amount or throughput (= traffic amount per unit time) is evaluated by calculating the delay time in units of clusters limited to the uplink.
That is, by calculating the delay time of messages sent and received between nodes according to network information (network configuration information, route management information, queue management information) managed in units of clusters, traffic volume or throughput for each cluster (= Traffic amount per unit time) is evaluated.

  A network simulation apparatus according to a second aspect is directed to the bit torrent protocol according to the first aspect, and each node in the cluster is a tracker or a peer that operates in an autonomous distributed manner.

  According to the present invention, in realizing a network simulation of P2P relay type content distribution, for a plurality of clusters to which a plurality of nodes are connected, a branch cluster that performs relay and a leaf cluster that is connected to only one branch cluster. Dividing and handling only the data sent by the nodes in the cluster in the inter-node link and inter-cluster link queues in consideration of the order waiting, so that the part that is faithfully reproduced as the network is the uplink of the access network It is limited to. As a result, the P2P relay type content distribution simulation can be realized with a small amount of calculation compared to the conventional network simulation apparatus.

  The conventional network simulation apparatus has a problem that it is difficult to simultaneously perform network simulation and autonomous distributed system simulation using a standard server PC from the viewpoint of computational complexity. According to the simulation apparatus, network information (network configuration information, route management information, queue management information) that is divided into a branch cluster that performs relay and a leaf cluster that is connected to only one branch cluster, and is set for each cluster. Accordingly, the delay time of messages transmitted and received between nodes can be estimated, so that both functions described above can be provided even when a standard server PC is used.

It is a functional block diagram for demonstrating the structure of the network simulation apparatus of this invention. It is a schematic diagram of the network topology to which the network simulation apparatus of the present invention is applied. (A) And (b) is a schematic diagram which shows the concept of the link in the network simulation apparatus of this invention. (A)-(c) is a schematic diagram for demonstrating the definition of the delay in the network simulation apparatus of this invention. (A)-(b) is a schematic diagram for demonstrating the generation | occurrence | production of a delay in the network simulation apparatus of this invention. It is a flowchart for demonstrating operation | movement in the network simulation apparatus of this invention. (A)-(e) is a schematic diagram which shows the example of a message handled with the network simulation apparatus of this invention. It is a flowchart explaining the outline | summary about the BitTorrent protocol utilized as a protocol for implement | achieving a P2P relay type | mold delivery system.

Hereinafter, an exemplary embodiment of the present invention will be described with reference to FIGS. 1 and 2.
The simulation apparatus 30 of the present invention calculates the traffic volume, throughput, and download time between nodes for a network that constructs a P2P relay system configured by connecting a large number of nodes (generic names of terminals, routers, servers, etc.). As shown in FIG. 1, it is configured to be connectable to a user 40 who performs the simulation in the network.

  The network simulation device 30 manages a network state, a simulation execution management unit 31 that executes a simulation according to a simulation operation flow described later in response to a request from the user 40, a setting management unit 32 that manages simulation setting information, and the network state. In response to a request from the network status management unit 33, a plurality of node execution units 34 that execute events in each node, a log management unit 35 that records the execution results of events in each node execution unit 34, and a user 40 A simulation result display unit 36 that displays a simulation result with reference to the log management unit 35 is provided.

The network state management unit 33 manages the network configuration simulated by the network simulation device 30, manages paths between clusters in the network (management of paths between clusters derived by SPF), and links between clusters described later. And queue management for the output queue of the node link. The network topology simulated by the network simulation device 30 is divided into clusters, which are several small networks, as shown in FIG.
Each cluster is divided into a branch cluster 10 which is a part for connection with other regions and a relay with an ISP, and a leaf cluster 20 which is a part from an access network to which a user (referred to as a peer in the P2P relay system) belongs, Each is connected to each other.

  The branch cluster 10 has a plurality of network relay devices (routers or the like) 11 connected to each other. Further, depending on the branch cluster 10, a server 12 {content server, distribution control server (tracker), etc.} serving as a part of the P2P relay system is set.

  The leaf cluster 20 is configured to include devices that aggregate access networks, and a plurality of peers 21 are set under the devices. The branch clusters 10 can be connected to each other without limitation, but the leaf cluster 20 is connected to only one of the branch clusters 10.

  The network simulation by the network simulation apparatus is defined in a simplified manner as follows. The route calculation is performed on a cluster basis for each branch cluster 10 or leaf cluster 20. That is, assuming that there is a logical link between clusters, managing the cost and propagation delay for each link is performed.

  The route from one cluster to another is calculated in the same way as the SPF algorithm used in OSPF. Therefore, when a node in the cluster communicates with a node in another cluster, communication is performed using the route calculated for each link by the above means.

  As shown in FIG. 3, the types of links include inter-cluster links that exist logically as links connecting clusters (link 15 between branch clusters 10 and link between branch cluster 10 and leaf cluster 20). 16) and a node link 22 that exists for each node existing in the leaf cluster 20 and connects the node to the network. Therefore, communication between a node belonging to one branch cluster 10 and a node belonging to another branch cluster 10 is shared by one intercluster link 15. Further, in the case of connection between a node belonging to the leaf cluster 20 and a node belonging to the branch cluster 20, a node link 22 in which the bandwidth is shared in units of nodes, and an intercluster link (link 16 between the branch cluster 10 and the leaf cluster 20). The communication between the nodes is performed via these two.

The setting management unit 32 manages simulation setting information when a network is simulated. The amount of traffic between clusters (nodes) is calculated at the time of simulation execution and accumulated in the log management unit 35. Thereby, the user 40 can acquire the traffic volume between specific clusters (nodes) after simulation execution.
The initial setting information of the simulation setting information is set by the user 40 before the simulation is executed, and includes a network configuration (branch cluster 10 or leaf cluster 20), a link connection configuration (inter-cluster link (link 15 and link 16). ) Or node link 22), information such as the number and type of nodes (peer or tracker) on each cluster, and node activation timing.

  The node execution unit 34 executes the operation of one node for the event managed in the event management according to the algorithm of either the peer or the tracker. In the network simulation apparatus of the present invention, since it is assumed that 10,000 or more independent nodes are simulated, there are as many node execution units 34 as there are independent nodes. Each node execution unit 34 is called by the simulation execution management unit 31 in the order of the node ID once in units of simulation processing time (SUT), and the simulation processing time (SUT) among the events managed in event management. Execute the event to be executed. Nodes have either peer or tracker algorithms and execute events according to their respective algorithms. Event execution results are recorded in chronological order and accumulated in the log management unit 35.

  The log management unit 35 can obtain a simulation result such as a traffic amount by storing event execution time, node, destination node, and event type information necessary for processing for each event.

The event processing in the node execution unit 34 will be described.
In the reception event process, the node state is updated as the message is received. The log management unit 35 is notified that reception event processing has been performed. Further, according to the peer / tracker algorithm, when it is necessary to transmit a message to another destination node (can be plural), a transmission event (can be plural) is set and transmission event processing is performed. The node state associated with the message transmission is updated, and the log management unit 35 is notified that the transmission event processing has been performed. Further, the network state management unit 33 is requested to calculate the delay from the corresponding node to the partner node, and as a result, the time when the delay time is added to the current simulation processing time (SUT) to the partner node is set as the event reception time. Set the reception event related to this message.

The simulation result display unit 36 refers to the log management unit 35 based on a request from the user 40 and displays the traffic amount and throughput as a simulation result.
This traffic volume is calculated as the total message length. The calculation timing is performed, for example, when the node transmits a message as a transmission event. The length of the message can be changed according to the simulation initial setting, but is fixed for each message type. For example, in the case of a piece message, it is set to 256 kilobytes. Basically, the message length is determined according to the BitTorrent protocol.
Throughput is the amount of traffic per hour, and is calculated by dividing the total message length by the total simulation target time.

Next, specific examples of simulation results will be described.
If the simulation processing time (SUT) of one unit is 100 milliseconds and 10,000 SUT is consumed in the entire simulation, the simulation is performed for 1,000 seconds. If the total message length between clusters A and B is 2,000 MB (megabytes), the traffic volume between clusters A and B is 2,000 MB, and the throughput between clusters A and B is ( 2,000 / 1,000) × 8 = 16 Mbit / sec.

Next, definitions of three types of delays (propagation delay, transmission delay, and queuing delay) that occur in data communication between nodes in the simulation of the network simulation apparatus will be described with reference to FIG. The unit of data handled by the network simulation apparatus is called a message.
The message corresponds to one to a plurality of IP packets as an IP protocol, but the network simulation device does not assume that the protocols such as IP, TCP, and UDP are strictly processed. Expressed in units of messages. The message includes a control message for the purpose of transmitting protocol control information and a data message for the purpose of transferring application data such as contents. In the network simulation apparatus of the present invention, it is assumed that the message size of the control message is zero.

As shown in FIG. 4A, the propagation delay is a delay required for a message to physically arrive in an intercluster link. This delay is related to the transmission of electrical signals and is proportional to the physical distance between branch clusters.
As shown in FIG. 4B, the transmission delay represents a delay required until transmission of a message is started on a link having a finite bandwidth and transmission is completed. The smaller the link bandwidth and the larger the message size, the greater the transmission delay.
As shown in FIG. 4C, the queuing delay is the time required from the occurrence of a message transfer request to the start of the corresponding message transfer process in a link sharing a plurality of message transfers. If the message cannot be transferred immediately, the message is accumulated in the output queue. This delay is not proportional to the size of the corresponding message, and increases as the output queue size increases and the link bandwidth decreases.

Next, a delay generation model using the above delay will be described with reference to FIG.
As shown in FIG. 5A, for a message transmitted by a node in the branch cluster 10, a queuing delay of the output queue for each branch cluster, a transmission delay and a propagation delay in the intercluster link 15 occur.
As shown in FIG. 5B, for messages transmitted by the nodes in the leaf cluster 20, queuing delay and transmission delay in the node link 22, and queuing delay, transmission delay and propagation in the intercluster link 16. There is a delay.
With respect to a message passing through a plurality of branch clusters 10 as shown in FIG. 5C, a transmission delay and a propagation delay in the passing branch cluster 10 occur every time each branch cluster is passed.

In addition, the following conditions are applied to the delay generation model shown in FIG. 5 in order to simplify the model and reduce the amount of calculation.
(1) There is no queuing delay in the intercluster link to be relayed or the downlink node link.
(2) The size of the output queue is considered infinite, and packet loss due to queue overflow does not occur.
(3) No data transfer error (bit error) occurs in the relay device or link.
(4) Therefore, no packet loss occurs on the network, and delay due to retransmission in TCP is not considered.
(5) The delay until the node in the branch cluster transfers the message to the router having the intercluster link is not considered.

The delay calculation formula can be displayed as a delay time De2e (seconds) due to end-ends of the node a belonging to the cluster X1 and the node b belonging to the cluster Xn, where X1 · X2... Xn is the cluster route. The delay time De2e is expressed as a function of Dp which is the propagation delay (second) of the intercluster link, Dt which is the transmission delay (second) of the intercluster link, and Dq which is the queuing delay (second) of the intercluster link. can do.
That is, Dp (XY) is the propagation delay (seconds) of the intercluster link between the clusters XY, B (XY) is the bandwidth (bps) in the XY direction of the intercluster link between the clusters XY, and M is When the message length (byte) and Qt-1 (XY) are the output queue length (byte) at the time t-1 of the intercluster link between the clusters XY, the propagation delay Dp and transmission in the intercluster links 15 and 16 are transmitted. The delay Dt and the queuing delay Dq are defined as shown in Table 1, respectively. Further, the delay time De2e is calculated by the equation 1 when the cluster X is a branch cluster.

  Further, B (a) is the upstream bandwidth (bps) of the node link of node a in the cluster X, and Qt-1 (a) is the time t- of the uplink of the node link of node a in the cluster X. Assuming that the output queue length (byte) is 1, the propagation delay Dp, transmission delay Dt, and queuing delay Dq in the node link 22 are defined as shown in Table 2, respectively. Further, when the cluster X is a leaf cluster, the delay time De2e (second) is calculated by Equation 2.

Next, the operation of the simulation execution management unit 31 will be described with reference to FIG.
In this simulation apparatus, each node executes an event-driven process independently for each simulation processing time (SUT). When the simulation apparatus is activated, the initial state of each node is set according to the preset items, and the SUT is set to the initial state “0” (step S101). For each SUT, the reception event at each node is processed (step S102), and then the transmission event is processed (step S103).
0 When there are messages sent and received by SUT, it is necessary to process the same node multiple times in the same SUT. For this reason, a rescan flag is provided, and when there is a rescan (step S104), the process of each node is repeatedly performed to increase the SUT.
When there is no rescanning, the SUT is increased (step S105), and the process in the next SUT is executed. When the simulation termination condition is satisfied (step S106), the result is output (step S107), and the simulation is terminated.

An event handled by the simulation apparatus will be described.
The event for each node is held in the event queue together with the time of occurrence, destination, event type, and parameters. Occurrence time is expressed as an elapsed SUT since the start of simulation. The partner is a number that identifies the partner node that transmits and receives the event. The event type is a message type or the like. The parameter is a parameter associated with the event. Messages transferred between nodes are processed as transmission / reception events in the transmission / reception nodes, respectively.

FIG. 7 shows an example of a message targeted by the network simulation apparatus in realizing the P2P relay type content distribution protocol.
FIG. 7A shows a case where message transfer is performed between nodes (N1 and N2) without a time delay.
FIG. 7B is an example in which a delay of SUT time of a time occurs in message transfer between nodes (N1 and N2).
FIG. 7C shows an example in which a transmission message is generated as a response to a received message in message transfer between nodes (N1 and N2). It is assumed that there is no processing delay in the system.
FIG. 7D shows an example in which a plurality of messages are transmitted from the node (N1) to the plurality of nodes (N2, N3) at the same time.
FIG. 7E shows an example in which a plurality of messages are exchanged within a 0 SUT between nodes (N1 and N2).

Next, an outline of a bit torrent protocol that is widely used as a protocol for realizing the P2P relay distribution system will be described with reference to FIG.
With BitTorrent, one file is divided into a plurality of pieces of fixed length (Piece), and Piece messages are taken as one unit to receive pieces that the player does not own independently from a plurality of peers. Each peer performs an autonomous decision for each peer regarding the peer to be transferred (Unchoke), the order of downloading pieces, and the like.

BitTorrent is composed of two types of nodes: a tracker that controls distribution and a peer that downloads / uploads content. Peer # 1 (peer # 1) that performs download first requests content from the tracker (step 201), and the tracker returns a peer list from which content can be downloaded to peer # 1 (peer # 1) ( Step 202).
Peer # 1 (peer # 1) performs TCP connection with each peer (peer # 2, peer # 3) notified in the peer list (steps 203, 205, 206 and steps 204, 207, 208). After that, connection (Handshake) at the BitTorrent level is performed (Steps 209 and 210 and Steps 211 and 212).

  Thereafter, for example, exchange of a piece map with respect to peer # 2 (peer # 2) (Bitfield) (step 213 and step 214), notification of interest to the peer (Interested / Not interested) (step 215 and step 216) Then, notification (Unchoke / Choke) (step 217 and step 218) of whether transfer is possible or not is performed. Thereafter, peer # 1 (peer # 1) requests a piece from peer # 2 (peer # 2) (step 219), and receives a piece (Piece) as a response (step 220).

Next, a method for realizing the above-described BitTorrent protocol with a network simulation apparatus will be described below. Piece messages are treated as data messages, and other messages are treated as control messages.
Message transfer is realized by setting a transmission / reception event for each message type in the network simulation apparatus. For the autonomous judgments made by each peer, the network simulation device has the same algorithm as BitTorrent.

In other words, in the peer algorithm, the connection status with the peer (TCP connection, choke / unchoke status, etc.), the piece held by the peer (part of the content), and information indicating the performance of the peer (download) Time) and the like, and from which peer peer, which piece is to be obtained, the timing for changing the choke / unchoke state, etc. are determined.
In the tracker algorithm, information on all peers that have requested download of a certain content is managed, and when a peer list request is received from a new peer, a certain number of peer lists are returned.
As algorithms, there are a Rarest first algorithm that preferentially downloads pieces with low distribution, and a Tit-for-tat algorithm that preferentially allows data transfer to peers with a high download speed.

Examples of simulation execution conditions are shown below.
A cluster, start time, and end time are set for each peer. The start time is set as the SUT time, and the end time is set such that n SUT time elapses after the download is completed. In order to simulate the statistically multiplexed behavior of multiple peers, it is possible to group settings for multiple peers. For example, it is possible to set in consideration of statistical behavior, such as starting at a fixed interval, completing after a period of time following a normal distribution in which a is an average and b is a variance after completion of download.
As the simulation result, the download time, the number of messages for each message type, and the like are output. Information necessary to output the results during node processing in each SUT is recorded and collected simultaneously after the simulation.

According to the network simulation device described above, since the basic design is to simulate each node in an event-driven manner using the simulation processing time (SUT) as a unit, simulation can be performed by changing the time of one unit of simulation processing time (SUT). There is an advantage that the accuracy of can be changed.
In addition, the impact on software implementation accompanying the increase in the number of nodes is small, and if high-performance hardware is introduced, the simulation time can be simply reduced.
Furthermore, since there are many parts that can be processed independently in units of nodes, it is easy to cope with multi-CPU and multi-threading.

  The network simulation apparatus described above divides the roles of the relay network and the access network, and delays messages sent and received between nodes according to network information (network configuration information, path management information, queue management information) managed in units of clusters. By calculating the time, traffic volume or throughput between clusters (= traffic volume per unit time) is evaluated, enabling simulation of a large-scale network with a small amount of calculation, and a scale equivalent to 10,000 to 10,000,000 It is suitable for evaluating a P2P relay type content distribution model for users.

  DESCRIPTION OF SYMBOLS 1 ... Network, 10 ... Branch cluster, 11 ... Network relay apparatus (router), 12 ... Server, 15 ... Intercluster link, 16 ... Intercluster link, 20 ... Leaf cluster, 21 ... Peer, 22 ... Node link, 30 ... Network simulation device 31 ... Simulation execution management unit 32 ... Setting management unit 33 ... Network state management unit 34 ... Node execution unit 35 ... Log management unit 36 ... Simulation result display unit 40 ... User

Claims (2)

In a network having a plurality of clusters (small networks) to which a plurality of nodes are connected,
The plurality of clusters are divided into a branch cluster for relaying and a leaf cluster connected to only one branch cluster,
The intercluster link between the branch clusters and between the branch cluster and the leaf cluster is shared by communication between nodes,
Occupied bandwidth for each node for the inter-node link between the nodes in the leaf cluster,
Each node has both an inter-node link and an inter-cluster link queue, and only the data transmitted by the nodes in the cluster is output in consideration of waiting in the inter-node link and inter-cluster link queues. A network simulation apparatus characterized in that a traffic amount or a throughput is evaluated by calculating a delay time for each cluster limited to a line.   The network simulation apparatus according to claim 1, wherein the node is a tracker or a peer that operates in a distributed manner, and is intended for a BitTorrent protocol.

Images