JP6076281B2 - ISP (Internet Services Provider) Revenue Forecasting Apparatus, Method, and Program - Google Patents

Description

  The present invention relates to an ISP (Internet Services Provider) revenue prediction apparatus, method, and program, and more particularly to the introduction of CCN (Content Centric Networking) that caches content in a router, and finds and receives content addressed to the content name. Accordingly, the exchange pattern of traffic between ISPs changes, but the revenue of ISPs that have introduced CCN to predict the impact of CCN introduction on ISP revenues using a hierarchical AS (Autonomous System) topology model. The present invention relates to a prediction apparatus, method, and program.

  On the Internet, the distribution ratio of various contents such as Web contents, UGC (User Generated Content) represented by YouTube (registered trademark), and large-capacity rich contents such as movies and dramas created by content providers (CP) large. Since packets are routed using an IP address in the conventional Internet, overhead is generated to resolve the IP address of the server serving as the distribution source from the name of the content prior to distribution.

  Therefore, CCN that caches content with a router in the network (NW) and routes a distribution request to a distribution server using the name of the content is attracting attention as an efficient content distribution NW. In CCN, a user who wants to acquire content sends Interest to the original holding server, and the router transfers Interest using the name of the content. The router is equipped with a memory for caching content (actually cached in units of chunks in which content is divided into multiple parts, but is referred to as content in this specification). The router caches the transferred content. If the router on the Interest path caches the requested content, it distributes the requested content to the request source without forwarding the Interest to the next hop router. By using CCN, the overhead of name resolution processing can be avoided, and content can be distributed from a position closer to the user than when distributing from the original server or cache server, and the delay time and NW load can be expected to be reduced.

  The Internet is configured by interconnecting multiple ASs that are autonomously operated by different organizations, but each NW is connected to a transit ISP and pays transit costs to connect to all NWs. (There are other forms of peering connection and IXP (Internet Exchange Point) connection where two or more ISPs exchange traffic with each other free of charge). Transit costs are generally metered according to 95% of the average data transfer rate every 5 minutes. In the current Internet, CDN (Content Delivery Network) is widely used to deliver content efficiently. However, since the traffic traffic pattern between ISPs changes with the introduction of CCN, transit costs between ISPs and ISPs Revenue will be affected. Since the introduction of CCN is based on the judgment of each ISP, in order to clarify the dissemination potential of CCN, it is necessary to analyze the impact of CCN introduction on the profit of each ISP.

  Refer to Non-Patent Document 1 below for the above technique.

B. Ahlgren, et al., "A Survey of Information-Centric Networking," IEEE Commun. Mag., Vol.50, no.7, pp.26-36, July 2012. H. Chang, S. Jamin, and W. Willinger, "To Peer or not to Peer: Modeling the Evolution of the Internet's AS-level Topology," IEEE INFOCOM 2006. DePeering International, http://depeering.net

  As a study that analyzed the actions autonomously taken by each AS in CCN, it was suggested that the motivation to cache content in each NW does not work in CCN, and the need to give incentives Some have analyzed the impact of the routing policy between ASs due to the introduction of CCN, but so far quantitatively analyzed the impact of the introduction of CCN on ISP revenue Is not seen.

  The present invention has been made in view of the above points. In a content-oriented network in which content is cached in a network device (router) and data is transferred with the content name as an address, a topology model between ASs is used between ISPs. It is an object of the present invention to provide an ISP revenue forecasting apparatus, method, and program capable of calculating an exchange pattern of traffic and estimating the impact on the network operator's revenue, such as a change in transit costs between ISPs. And

According to one aspect, in CCN (Content Centric Networking) that caches content in a router and discovers and receives content addressed to the content name, predicting the impact of the CCN introduction on ISP (Internet Services Provider) revenue ISP revenue forecasting device for
Using AS topology data and data that classifies the attributes of each AS, AS that has no p2c link (provider-to-customer) and c2p link (customer-to-provider), and this p2c link And the AS that provides transit service to other ASs, and the AS that excludes the AS that has the c2p link for these ASs are extracted, and the c2p link is extracted from the extracted ASs. An AS that does not have at all is extracted and assigned to Layer 1, and then an AS that has one or more c2p links is extracted from the ASs that are not classified into Layer 1 with respect to the AS classified as Layer 1 In the same manner, the process of extracting an AS having one or more c2p links and assigning a layer k + 1 to an AS classified as layer k is assigned to an integer greater than or equal to k = 2. Ascending order until all ASs are assigned layers , the AS connection structure model that models the connection structure between the ASs is repeated. Dellization means,
Modeling the content, calculating the transit cost according to the data transfer rate that flows on the transit link, applying the transit cost to the given delivery cost, determining the network cost, and the cost of the cache memory according to the capacity Cost model generation means for calculating
Based on the connection structure between the ASs modeled by the AS connection structure modeling unit, the revenue of each AS is calculated using various costs calculated by the cost model generation unit, and the revenue of each AS is calculated. ISP revenue calculation means for calculating the ISP revenue;
An ISP revenue forecasting apparatus is provided.

  According to one aspect, with the introduction of CCN that caches content at a router and discovers and receives content addressed to the content name, the traffic exchange pattern between ISPs changes. It can be used to predict the impact of CCN adoption on ISP revenue.

The structural example of the ISP profit prediction apparatus in one embodiment of this invention. The connection configuration model between AS (Autonomous System) in one embodiment of this invention. The example of the path | route of the content delivery flow in one embodiment of this invention. An example of interest and a content delivery route when applied to CCN in one embodiment of the present invention. The example of the monthly profit of each AS in one embodiment of the present invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  The present invention models the AS, estimates the AC traffic of the modeled NW, calculates the transit cost exchanged between ISPs from the estimated traffic volume, further calculates the network cost, the cost of the cache memory, The revenue of each AS is calculated from these costs.

  FIG. 1 shows a configuration example of a profit prediction apparatus according to an embodiment of the present invention.

  The revenue prediction apparatus shown in the figure includes an AS connection structure modeling unit 101, a cost model generation unit 102, an ISP revenue calculation unit 103, and an ISP revenue output unit 104.

  The AS connection structure modeling unit 101 models a connection structure between ASs. The cost model generation unit 102 generates various cost models. The ISP revenue calculation unit 103 calculates the ISP revenue using the modeled AS connection structure and cost model. The ISP revenue output unit 104 outputs the ISP revenue derived by the ISP revenue calculation unit 103.

  Next, details of the revenue prediction method of the ISP introducing the CCN according to the embodiment of the present invention will be described.

[1] Assumptions:
This section describes various preconditions necessary for deriving exchange traffic between ASs, such as distribution request generation patterns and inter-AS topology configuration.

  The following content modeling, transit cost, network (NW) cost, and cache cost are set by the cost model generation unit 102 of FIG. 1 and stored in a memory (not shown).

<Content modeling>
Assume an environment where large-capacity M rich contents such as movies and dramas are provided on all AS NWs. For simplicity, the size of each content is assumed to be uniform and L (Mbyte). In addition, it is assumed that the average number of viewing times per month of each user is d, and that each content m is selected with a certain probability q _m in each distribution request. Assume that the content IDs are assigned in descending order of q _m , where q ₁ is the content ratio for the most popular content and q _M is the content ratio for the least popular content. Q (m) is the cumulative distribution of q _m

. The total number of users accommodated by all ASs is U ₁ , and the total number of CPs is U ₂ .

<Transit fee>
Currently, many ISPs charge customers ISPs with a transit contract according to the data transfer rate that flows on the transit link during the month. According to the analysis of transit costs used by 20 regional ISPs in the US in 2004, the monthly transit cost T is proportional to the 0.75th power of the data transfer rate V (Mbps) (see Non-Patent Document 2), T = 100V ^0.75 Can be approximated by

Many ISPs use 95% of the data transfer rate every 5 minutes as V, but assume that the 95% value is three times the average transfer rate. On the transit link, traffic flows in the downhill direction from the provider to the customer and the uphill direction from the customer to the provider. However, when V is based on the value in the large direction (max model), the total value in both directions There are cases where transit costs are calculated based on the sum model. Therefore delivery count is D _d of the contents flowing through the monthly transit link in the direction of each of the downhill and uphill, when a D _u, the values of V used transit cost calculation When 30 days the number of days of one month
V = (3 × 8LD) / (30 × 24 × 3600) = 1.08 × 10 ^-5 LD
It becomes. Where D is

The monthly transit cost T is given by

<Network (NW) cost>
Each ISP needs to bear the NW construction and operation costs, but it is assumed that a distribution cost of κ will be incurred each time content is distributed on its own network, and the monthly transit costs will be used as the distribution cost. Apply. In 2013, the average transit price per Mbps in the US is 1.57 USD (see Non-Patent Document 3), and the peak demand for commercial VoD (Video on Demand) service is about 1.8 times the average. Has been reported,
κ = (1.57 × 1.8 × 8L) / (30 × 24 × 3600) = 8.72 × 10 ^-6 L
Give in.

<Cash cost>
When an ISP uses a CCN, it is necessary to introduce a cache memory in each router, so it is necessary to bear the cost of the cache memory. In this specification, assuming that DRAM (Dynamic Random Access Memory) is used and that the cache memory is replaced with 3 years, and the memory cost per 1Mbyte is 0.016 USD, the capacity is B (unit is the number of contents). The cost of memory is
0.016LB / 36 = 4.44 × 10 ^-4 LB (USD)
It becomes.

[2] AS modeling:
The AS modeling is performed by the AS connection structure modeling unit 101 in FIG.

  In this section, the inter-AS topology, CP / user accommodation pattern, and cache design assumed in this specification are described.

<Inter-AS topology>
The form when two ASs are connected can be classified into a transit connection in which one collects transit costs according to the amount of data transferred from the other, and a peering connection in which both do not pay each other for a fee. If the two ASs differ greatly in size, the larger one becomes a provider and provides transit service to the other AS. The other AS becomes a customer and pays transit costs to the provider. On the other hand, peering connections that do not generate transit costs are often used when ASs of the same scale are connected.

If two ASs, x and y, are transited as x is provider and y is customer, the link that connects them as seen from x is the provider-to-customer (p2c) link and y It is called a customer-to-provider (c2p) link. Thus, the same transit connection link is a p2c link for one and a c2p link for the other. On the other hand, a link connecting two ASs by peering connection is called a peer-to-peer (p2p) link. If the AS of layer k (hereinafter referred to as L _k ASL) x has a p2c link to L _{k + 1} ASy, then y will have a c2p link to x, for one of the ASs at both ends It is a p2c link, and for the other AS, the same link is a c2p link.

  In this specification, the AS topology is modeled by using the following two data relating to the topology between ASes disclosed on the CAIDA website (http://www.caida.org/data).

As_rel file:
`` As_rel file '' is 18,967 ASs using BGP (Border Gateway Protocol) table information (RouteView) and inter-AS topology data estimated in 2005 from inter-AS routing policy information (IRR: Internet Routing Registries). This is data obtained by classifying 85,136 links existing in between into p2p, p2c, and c2p.

As2attr file:
“As2attr file” is data obtained by classifying the attributes of each of 19,537 ASs into Large ISPs, Small ISPs, Universities, IXPs, NICs, and Customers using the inter-AS topology data.

  The AS connection structure modeling unit 101 uses the above two data to model the topology between ASs in a tree-type hierarchical topology according to the procedure described below.

  (1) Among the ASs included in both data files, the p2c link and the c2p link are selected from 17,828 ASs classified as Large ISPs, Small ISPs, Universities, or Customers in the as2attr file. Three ASs that do not have any, 481 ASs that have p2c links and provide transit service to other ASs, and 201 ASs that have c2p links to these 481 ASs Excludes 17,143 AS.

  (2) From the extracted 17,143 ASs, an AS that does not have any c2p link (has not purchased transit from any ISP) is extracted and assigned to Layer 1.

  (3) From the ASs not classified into layers, ASs having one or more c2p links are extracted from the ASs classified as Layer 1 and assigned to Layer 2.

  (4) After that, in the same way, the process of extracting ASs having one or more c2p links for ASs classified as layer k and assigning layer k + 1 is performed in ascending order for integers greater than or equal to k = 2. Repeat until all 17,143 ASs are assigned layers.

  As a result of the above processing, the AS of 17,143 is classified into 7 layers. As an example, the number of ASs classified in each layer is summarized in Table 1 for each AS attribute in the as2attr file.

Universities AS and Customers AS can be regarded as NWs that accommodate users and CPs. Since the purpose of this specification is to analyze transit costs, only Large ISPs and Small ISPs are considered as the AS of the AS topology. Therefore, AS and ISP are used interchangeably. The above (2) to (4) are repeated for 5,473 ASs classified as Large ISPs or Small ISPs out of 17,143 ASs, and the number of ASs classified when each layer is assigned is shown. It is shown in 2.

Since most large ISPs and small ISPs are assigned to any one of layers 1 to 3, the inter-AS topology is modeled using 4,737 ASs classified into these three layers. That is, the number of layers is K = 3, and the number of ASs of each layer k is N ₁ = 49, N ₂ = 2,123, and N ₃ = 2,565. And

Are defined as the average number of p2c links for L _{k + 1} AS for each L _k AS, the average number of c2p links for L _k-1 AS, and the average number of p2p links for L _k AS. Calculated from the connection structure of 4,737 AS in Table 3

Summarize the values of.

FIG. 2 shows a connection configuration model between ASs assumed in this specification.

Each L _k AS (1 ≦ k ≦ K-1) is equal to each L _{k + 1} AS

Have a p2c link, and each L _k AS (2 ≦ k ≦ K) is equally probable with each L _k-1 AS

It is assumed that the network has a c2p link (the c2p link does not exist in the AS of layer 1 and the p2c link does not exist in the AS of layer K = 3). Therefore, for each k of 1 ≦ k ≦ K-1,

(The values shown in Table 3 are rounded to the second decimal place, so the value calculated from this value has a slight error). In addition, each L _k AS of each k (1 ≦ k ≦ K) is uniform with other ASs belonging to the same layer k.

Suppose you have a p2p link.

<CP / User accommodation pattern>
Next, consider the CP that provides the content and the user that requests the content. In this specification, for the sake of simplicity, it is assumed that CPs and users are accommodated uniformly for ASs classified as Universities or Customers. In Table 1, since the Universities AS or Customers AS assigned to layer k has a c2p link to L _k-1 AS, the ratio W _k between CP and user accommodated in L _k AS is W _1. = 0.460, W ₂ = 0.426, W ₃ = 0.114.

In each layer k, it is assumed that each _Nk AS uniformly accommodates CPs and users, and the request ratio for the contents provided by each CP, the distribution request generation rate of each user, and the selection probability of each content Assumes uniformity.

At this time, when a content distribution request occurs, the request source user and the probability ω _k that the CP holding the original of the requested content is accommodated in each L _k AS is

It becomes.

<Cache design>
Since CCN has a cache installed at each router, the AC traffic between ASs is affected by the topology in AS, the cache capacity of each router, cache operation policy, and so on. However, for the sake of simplicity in this specification, each AS that has introduced CCN has arrived while avoiding duplication of cache contents in each router by exchanging cache information between all routers in its own network. If any router has content for Interest, Interest is transferred to that router, and the content is transmitted to the request source. Therefore, only the total cache capacity B of all routers in each AS is considered (unit is the number of contents) without considering the cache capacity of each router and the topology within the AS. All ASs assigned to the same layer have the same B, and the cache capacity of each L _k AS is B _k . Since AS is expected to be larger in ASs belonging to higher layers, B _k > B _{k + 1} (1 ≦ k ≦ K−1) is assumed.

[3] AS policy assumptions
BGP is used for routing between ASs, but advertisement judgments on neighboring ASs in route information and route selection depend on AS policies that act autonomously. In addition, the autonomous decision of each AS determines which CCN is introduced, which content is cached, and which cached content is advertised to the adjacent AS. As a result, traffic exchanges between ASs strongly depend on various AS policies. Therefore, in order to analyze the impact of CCN on transit costs, it is necessary to organize the assumed AS policies. In this specification, the policies listed below are assumed.

<Route advertisement between AS>
Since provider AS is obligated to provide reachability (transit) for the entire Internet to customer AS, it advertises all route information received from all neighboring ASs. Also, since two ASs in peering connection can usually freely transmit and receive each other's traffic free of charge, each AS advertises all the received routes to the p2p connected AS. On the other hand, when the customer AS advertises the route to the provider AS, the traffic amount in the downhill direction on the c2p link increases, so the route is not advertised to the provider AS. However, since there is an obligation to guarantee the reachability of the users and CP's original server to the Internet that all ASs (defined as customer cones) that can be reached via the p2c link only from themselves, these addresses are not included. On the other hand, it advertises to provider AS. Since the uppermost L ₁ AS needs to ensure reachability to the Internet for its own customers, routing information is completely exchanged between all L ₁ ASs.

<Routing between AS>
When each AS receives an Interest for an address for which a route advertisement is received from multiple peer ASs of the p2p connection destination, p2c connection destination customer AS, and c2p connection destination provider AS, the p2c connection destination (transit cost) ), P2p connection destination (transit fee is not received / received), c2p connection destination (pay transit fee), Interest is transferred. As a result of this policy, when the uppermost layer via the AS route is k, only a c2p link is passed from the distribution server to the L _k AS, and a single AS or p2p link is used for layer k. After using multiple ASs, the content is delivered to the requesting user via only the p2c link. Such a characteristic of the path between ASs is known as Valley-free routing.

<AS cash judgment>
An AS using CCN can freely determine whether or not the content flowing in its own network is cached in the router. By caching the content that has been transferred in the downhill direction from the provider AS to the c2p link, when receiving the Interest for the same content from its own customer cone after the next time, without transferring the Interest to the provider AS, The corresponding content can be distributed from its own cache, and as a result, it is possible to reduce the amount of traffic flowing in the downhill direction through the c2p link. Therefore, the content received from provider AS is cached in its own network. On the other hand, if the content transferred from the customer AS in the uphill direction to the p2c link is cached, if the Interest for the same content arrives from the provider or peer AS, the content does not flow in the uphill direction on its own p2c link, and the customer AS Do not cash because transit income from Also, the content received from the peer AS is not cached in order to save cache resources because there is no transit cost for the traffic flowing on the p2p link.

[4] Inter-AS Traffic Volume This section describes the process of deriving the traffic volume generated on the connection link between ASs based on the above-mentioned assumptions in ISP revenue calculation unit 103.

<When not using a cache>
First, consider a case where there is no cache in all NWs and content is distributed from the original server for all distribution requests (WOC: without cache).

To the distribution request from the user received in the L _s AS, if the original server requests the content is accommodated in L _s AS, the data packets starting with the original server using only c2p links, 1 ≦ K It reaches u ₁ via only one AS _uk among AS belonging to each layer k of ≦ s. Then, the data packet is transferred to L ₁ ASd ₁ using the p2p link (d ₁ = u ₁ when returning with the same AS in layer ₁ ), and each layer k with 1 ≦ k ≦ r using only the p2c link The request source is reached via only one ASd _k among the ASs belonging to.

However at any t of 1 ≦ t ≦ min (r, s), when the ASu _t and Asd _t is peering connection, via the top layer of the delivery flow will be t. FIG. 3A shows an example in the case where t = 1 at r = 2 and s = 3, and FIG. 3B shows an example in the case where t = 2. Therefore, if the probability that the highest layer via the delivery flow is t is G _{r, s, t} , for each t of 1 ≦ t ≦ min (r, s),

Thus, G _{r, s, t} = 0 for each t where min (r, s) <t ≦ K. Therefore, for a given r and s, if the probability that the distribution flow passes through each c2p link of each L _k AS in the uphill direction is φ _{r, s, k} , when k ≦ s,

When k> s, φ _{r, s, k} = 0. Similarly, the probability that the distribution flow passes through each c2p link of each L _k AS in the downhill direction is φ _{r, s, k} because it is assumed that the CP and the user are arranged uniformly. Moreover, since the probability of being included in the customer cone of the L _k AS where the distribution requesting user or the original server is 1 / N _k each, each p2p link of each AS of layer k and each CP accommodated by itself Therefore, the probability μ _{r, s, k} that the distribution flow flows into the own NW is obtained by the following equation.

The distribution flow flows from each c2p link of each L _k AS to Fu _{, k as the} probability that the distribution flow passes in the uphill direction, F _{d, k as the} probability that the distribution flow passes to the downhill direction, from the peer AS or its own accommodation CP. If the probability is F _{p, k} ,

The probability H _k that the delivery flow flows into the NW of each L _k AS is

Given in. However, F _{u, K + 1} = 0.

<When using CDN>
Next, consider the case where all ASs have CDNs.

Each L _k AS has a cache of capacity B _k , and when receiving a distribution request from a user with a ratio ω _k that each accommodates, if the corresponding content exists in its own cache, it is distributed from the cache, If it does not exist, it is distributed from the original server. Therefore, the cache of each AS is used only for the accommodating user. Since each L _k AS does not cache the content of the CP that it contains, the probability that each content will be cached is 1−ω _k . Assuming cache replacement control using LRU (Least Recently Used), highly popular content is cached, so it is assumed that the top B _k / (1-ω _k ) popular content exists in the cache.

At this time, if the virtual cache upper limit ρ _r is defined as ρ _r = min {B _{r /} (1−ω _r ), M}, for the user accommodated in LrAS, the content m (1 ≦ m ≦ ρ _r ) is delivered from the cache. Therefore, the probability of distribution from the original server is 1−Q (ρ _r ), and the following equation is obtained.

H _k is also obtained by equation (8).

<When using CCN>
Next, consider the case where all ASs have CCN. As mentioned above, each L _k AS

In order to exchange route information to content existing in each customer cone and its own network with AS

It is possible to avoid a duplicate cache of the same content between ASs. Therefore, in each L _k AS, the probability that each content is cached is

And the virtual cache upper limit σ _k is

In other words, it can be considered that the content m in the range of 1 ≦ m ≦ σ _k is cached in each L _k AS.

Delivered on the c2p link of a certain L _k AS when the highest layer through which the Interest for the content stored in the L _s AS is sent from the user accommodated in the L _r AS is t The flow flows in the uphill direction because the Interest sent from the user who does not belong to the customer cone of x passes through each AS of layers r, r-1, ..., t, t + 1, ..., k-1 , X.

FIG. 4A illustrates an Interest and a content distribution route when the distribution flow passes through the c2p link of x in the uphill direction when r = 3, s = 3, and t = 1. Interest reaches x only when the content is not cached in all of ASa, b, c, and d via Interest. As described above, B _i <B _j is assumed for i <j, so the maximum value of σ _k on the Interest path is σ _t , and content in the range of 1 ≦ m ≦ σ _t is x Interest has not arrived.

On the other hand, the distribution flow on the c2p link of L _k ASx flows in the downhill direction because the Interest sent from the user belonging to the customer cone of x is the AS of each of the layers r, r−1,. In this case, it reaches the L _k-1 AS that is the provider of x via x, and the interest in the range of 1 ≦ m ≦ σ _k does not reach the L _k-1 AS. FIG. 4B shows an example. Interest reaches d only when content is not cached in ASe and x. Further, since the flow of distribution flows from the peering AS of x is not cached in the AS of the layer r, r−1,..., K + 1 that passes through, the content flow is limited to the content of m> σ _{k + 1} . Therefore, the following equation is obtained.

H _k can also be obtained from equation (8).

[5] Derivation of Monthly Revenue for Each AS The following is a process for calculating monthly revenue as the processing of the ISP revenue calculation unit 103. In the following description, the unit of revenue calculation is described as a month, but the period of the calculation unit is not limited to a month. The following transit costs, NW costs, and cash costs are values calculated by the cost model generation unit 102 and stored in a memory (not shown). The ISP revenue calculation unit 103 reads the values, and Perform the process.

Revenue R _k (USD) that each L _k AS earns in a month is

It is obtained by. However,

Is the transit cost each L _k AS receives from the customer AS of layer k + 1 per month,

Is the transit cost that each L _k AS pays monthly from the provider AS of layer k−1.

It becomes. However, F _k is defined by the following equation.

A _{r, k} is the monthly access cost that each L _k ASL obtains from its own accommodation user, and assuming that each user collects a fixed amount P _r (USD) fee monthly, A _{r, k} = P _r ω _k U ₁

Also, _{Ac, k} is the access cost that each L _k AS gets from its own accommodation CP per month, assuming a pay-as-you-go billing similar to transit costs between ISPs.
A _ck = 100 {1.08 × 10 ^-5 L _d U ₁ / U ₂ } ^0.75 U ₂ ω _k
It becomes.

Also, C _{n, k} is the NW cost that each L _k AS bears monthly, using the delivery cost κ,
C _{n, k} = κdU ₁ H _k
C _{m, k} is the cost of the cache memory (cache cost) that each L _k AS bears monthly.
C _{m, k} = 4.44 × 10 ^-4 LB _k
It becomes.

  A profit value is obtained from the cost calculated as described above and output to the profit output unit 104 of the ISP.

[6] Numerical evaluation <Evaluation conditions>
The monthly access fee collected from each user by the ISP is set to Pr = 50 USD, and the total number of users and CPs is set to U ₁ = 10 ⁹ and U ₂ = 10 ⁴ , respectively. In addition, the average number of times of viewing the content for each user is set to d = 10, and the average size of the content is set to L = 3 × 10 ⁴ Mbytes. Assuming that the total number of contents is M = 10 ⁶ and the required ratio q _m of each content m follows the Zipf distribution of parameter θ,

Set to. Also, the cache capacity of the lowest layer L _K AS in the CDN or CCN is set to B _K = B _min , and the parameter ε taking a real value of ε> 1 is used for each k of 1 ≦ k <K. Set _k = εB _{k + 1} . In the subsequent numerical evaluation, unless otherwise specified, θ = 1, B _min = 10, and ε = 5 are set.

<Monthly ISP revenue>
Fig. 5 shows that each L _k AS is obtained for each month when θ or B _min is changed for each of the three modes of WOC, CDN, and CCN and the two transit cost models of max model and sum model. Plot the revenue R _k . When sum is used, since the transit cost is higher than max, the revenue of each L ₁ AS is large, and the revenue of each L ₂ AS and L ₃ AS is small. CCN has an effect of reducing NW cost for L ₁ AS compared to CDN. However, in the case of sum, the decrease in transit costs received from L ₂ AS is larger than that, and CCN is smaller in revenue. On the other hand, in the case of max, the transit cost is determined by the downhill direction (F _{d, k} ) with a large amount of traffic, so the degree of reduction of the transit cost obtained from L ₂ AS due to CCN is small, and NW cost reduction effect and cancellation CCN and CDN The difference is small. As θ and B _min increase and the cash effect increases, the decrease in transit cost revenue for L ₁ AS exceeds the decrease in NW cost, but the revenue decreases, but for L ₂ AS and L ₃ AS, transit cost Profits will increase because both NW and NW costs decrease.

  It should be noted that the operation of the components of the ISP revenue prediction apparatus 100 can be constructed as a program, installed and executed on a computer used as the ISP revenue prediction apparatus, or distributed via a network.

  The present invention is not limited to the above-described embodiment, and various modifications and applications are possible within the scope of the claims.

100 ISP Revenue Forecasting Device 101 AS Connection Structure Modeling Unit 102 Cost Model Generation Unit 103 ISP Revenue Calculation Unit 104 ISP Revenue Output Unit

Claims (6)

In the CCN (Content Centric Networking) that caches content in the router and discovers and receives content addressed to the content name, ISP revenue prediction to predict the impact of the CCN introduction on ISP (Internet Services Provider) revenue A device,
Using AS topology data and data that classifies the attributes of each AS, AS that has no p2c link (provider-to-customer) and c2p link (customer-to-provider), and this p2c link And the AS that provides transit service to other ASs, and the AS that excludes the AS that has the c2p link for these ASs are extracted, and the c2p link is extracted from the extracted ASs. An AS that does not have at all is extracted and assigned to Layer 1, and then an AS that has one or more c2p links is extracted from the ASs that are not classified into Layer 1 with respect to the AS classified as Layer 1 In the same manner, the process of extracting an AS having one or more c2p links and assigning a layer k + 1 to an AS classified as layer k is assigned to an integer greater than or equal to k = 2. Ascending order until all ASs are assigned layers , the AS connection structure model that models the connection structure between the ASs is repeated. Dellization means,
Modeling the content, calculating the transit cost according to the data transfer rate that flows on the transit link, applying the transit cost to the given delivery cost, determining the network cost, and the cost of the cache memory according to the capacity Cost model generation means for calculating
Based on the connection structure between the ASs modeled by the AS connection structure modeling unit, the revenue of each AS is calculated using various costs calculated by the cost model generation unit, and the revenue of each AS is calculated. ISP revenue calculation means for calculating the ISP revenue;
ISP revenue forecasting device characterized by having. The ISP revenue calculation means is:
Transit costs that the layer k AS (L _k AS) receives from the layer k + 1 customer AS,
Transit costs each L _k AS pays the provider AS layer k-1 a, A _r, access costs each of _k L _k AS is obtained from its users accommodated, A _c, each L _k AS the _k is itself housed Access cost obtained from the content provider (CP), C _{n, k} is the network cost borne by each L _k AS, C _{m, k} is the cost of the cache memory borne by each L _k AS,
2. The ISP revenue prediction apparatus according to claim 1, further comprising means for calculating the revenue of each AS. The ISP revenue calculation means is:
The transit cost
When calculating
If there is no cache in all networks and the content is transmitted from the original server for all distribution requests, all the ASs will receive a CDN (Content Delivery Network). 3) The ISP revenue prediction apparatus according to claim 2, further comprising means for calculating every AS when CCN is introduced. In the CCN (Content Centric Networking) that caches the content in the router and discovers and receives the content addressed to the content name, the ISP that is executed by the device that predicts the impact of the CCN introduction on the revenue of the ISP (Internet Services Provider) A revenue forecasting method,
Using AS topology data and data that classifies the attributes of each AS, AS that has no p2c link (provider-to-customer) and c2p link (customer-to-provider), and this p2c link And the AS that provides transit service to other ASs, and the AS that excludes the AS that has the c2p link for these ASs are extracted, and the c2p link is extracted from the extracted ASs. An AS that does not have at all is extracted and assigned to Layer 1, and then an AS that has one or more c2p links is extracted from the ASs that are not classified into Layer 1 with respect to the AS classified as Layer 1 In the same manner, the process of extracting an AS having one or more c2p links and assigning a layer k + 1 to an AS classified as layer k is assigned to an integer greater than or equal to k = 2. Ascending order until all ASs are assigned layers , the AS connection structure model that models the connection structure between the ASs is repeated. Dellization step,
Modeling the content, calculating the transit cost according to the data transfer rate that flows on the transit link, applying the transit cost to the given delivery cost, determining the network cost, and the cost of the cache memory according to the capacity A cost model generation step for calculating
Based on the connection structure between the ASs modeled in the AS connection structure modeling step, the revenue of each AS is calculated using the various costs calculated in the cost model generation step, and from the revenue of each AS ISP revenue calculation step for obtaining the ISP revenue;
ISP revenue forecasting method characterized by In the ISP revenue calculation step,
Transit costs that the layer k AS (L _k AS) receives from the layer k + 1 customer AS,
Transit costs each L _k AS pays the provider AS layer k-1 a, A _r, access costs each of _k L _k AS is obtained from its users accommodated, A _c, each L _k AS the _k is itself housed Access cost obtained from the content provider (CP), C _{n, k} is the network cost borne by each L _k AS, C _{m, k} is the cost of the cache memory borne by each L _k AS,
The ISP revenue prediction method according to claim 4 , wherein the revenue of each AS is calculated by: Computer
An ISP revenue prediction program for causing the ISP revenue prediction apparatus according to any one of claims 1 to 3 to function as each unit.