Paper Title Here Hiroshi Fujinoki Department of Computer Science Southern Illinois University Edwardsville Edwardsville, IL 62026-1656 E-mail: hfujino@siue.edu Voice: +1-618-650-3727, Fax: +1-618-650-2555 Abstract In this paper, we present our new designs of interdomain SDN in the Internet to optimize the flow of network traffic by improving the utilization of the unused transmission bandwidth at tier-2 and 3 providers (the reality issues in the existing Internet). Address the problem of “beyond right” issue in the peering of tier-2 and 3 ISPs. A solution for the beyond right issue. Assessment the potential of the proposed solution. Categories and Subject Descriptors C.2.6 [Internetworking]: Routing protocols; C.2.6 [Internetworking]: Standards; Keywords Inter-domain routing, network traffic optimization, software defined networks, Border Gateway Protocol (BGP), routing in the Internet 1. Introduction (1) Why is “inter-domain routing” an important problem? Optimizing network traffic, such as optimizing network traffic load (i.e., transmission rate), transmission delay, and especially hardware resource utilization, has huge financial impact. (2) Why is inter-domain routing a difficult research problem? The structure in the Internet has continuously changed ever since its prototype was built in the late 60’s as ARPANET. Since ARPANET, the structure of the Internet had been quite much like a tree structure with the inter-connections of tier-1 ISP's at its root until the mid 90s. In the tree structure, network traffic generated by an Internet user first climbs up to a tier-1 ISP through a tier-2, where the traffic is switched to another tier-1 ISP. Then, it goes down to the destination through a tier-2 ISP. The recent popular adoptions of multi-homing and peering among tier-2 and even tier-3 ISP's have made the Internet more like a layered mesh structure [0]. The recent structural evolutions from a simple tree structure to a layered mesh structure has significantly increased opportunities for network traffic optimization [0]. Despite the increase in the inter-connectivity among tier-2 and 3 providers, network traffic flow between two domains in the Internet is still regulated based on their relationship, such as providercustomer, peer-to-peer, and sibling relationships [a]. Especially in peering tier-2 and 3 ISP’s, most of them do not offer beyond rights by not advertising their BGP paths to their peers because offering beyond rights to other peers will not directly contribute to own profit unless other incentives are provided. No beyond rights in tier-2 and 3 peering relationships cause sub-optimal utilization of available link bandwidth as follows. Figure 1 (a) shows an example of AS (autonomous system) interconnections, where AS 1, 2, and 3 are tier-1 ISPs and 4, 5, 6, 7 are tier-2 ISPs. The three tier-1 ISPs are fully connected, while AS 4 and 5, and 6 and 7 are connected by peer relationship. We defined “beyond rights” by the following two rules: (1) Non-tier1 ISPs: on an outgoing link at an AS if incoming network traffic is not destined to local customers (local customers include those who are downstream of directly connected intermediate ISPs). (2) Tier1 ISPs: since tier-1 ISPs do not pay to any other ISPs (a. k. a., “settlement-free”), it depends on each tier-1 ISP. If a tier-1 ISP announces a particular BGP path, any traffic flowing to its downstream is not considered a violation of beyond right principle. Otherwise, they follow rule (1) for non-tier1 ISPs. Based on the two rules, when user W transmits a payload traffic to Z, his traffic goes through AS 4, 1, 3, and 7. However, if AS-1 is down (or unavailable due to over traffic), W’s traffic can not go through link 5-2 and 2-3 (assuming that AS 2 does not advertise a BGP path of 2-3-7 to AS 1), because of “no beyond right” over AS 5 (Figure 1 (b)). AS 4 will not even see the path because AS 5 will not advertise its BGP path (5-2-3-7) to AS 4. - peer-peer - provider-customer - violation of “beyond right” - unavailable link 1 Tier-1 ISPs 2 1 3 2 3 Tier-2 ISPs 4 Users W 5 6 7 4 X Y Z W (a) AS-1 up 5 6 7 X Y Z (b) AS-1 down Figure 1 - an example of AS (autonomous system) interconnections Simply forcing AS 5 to announce its BGP path to reach AS 7 (5-2-3-7) to AS 4 may cause penalty from AS 2 for increased traffic load from AS 4 when any customers of AS 4 do not pay to AS 5. While interconnections of tier 2 and 3 ISPs by peering are expected to increase in the future [b], banned beyond rights among the tier-2 and 3 ISPs will significantly limit the opportunities offered by their peer-to-peer interconnections. The rest of this paper is organized as follows. Section 2 presents the major existing work. In Section 3, we present our design of ________ (using a mechanism of monetary insensitive to encourage beyond rights). Section 4 presents our assessments of how much the proposed ______ can contribute to utilize tier-2 and 3 inter-connections in the current Internet. Section 5 describes the conclusions and the planned future work, followed by references. 2. Existing Work BGP [0]. Inter-domain Traffic engineering (TE) [0, 0, 0, 0]. Traffic-load optimization between two neighboring ASs [c]. Inter-domain traffic engineering by GMPLS [d]. Inter-domain traffic engineering SDN [e]. E-payment systems [0]. Internet structure [0]. SDN is an emerging network traffic managing technique, which was initially developed for intradomain network traffic [0]. SDN provides a higher level of controls MPLS and over-lay networking do and it has several advantages: SDN provides logically centralized control over network traffic, which usually produces better results compared to a distributed approach. Because of its centralized nature, SDN does not require complex procedures to reflect the routing decisions to the controlled routers. For example, disseminating route updates has been a major issue in most of the distributed routing protocols, including the existing BGP, since their long convergence delay caused routing loops [0, 0, 0]. SDN is more flexible than MPLS and over-lay networking in that SDN can use more diverse input parameters in making routing decisions, even including' business logics each domain prefer [0]. (c) Electronic payment systems A taxonomy of electric payment systems 3. Proposed inter-domain SDN for optimizing network traffic in global Internet There are three major functional components in the proposed ____: disseminations of BGP path under beyond rights, signaling of path establishment, and traffic accounting. Disseminations of BGP path beyond rights: BGP paths are disseminated with an extension of the existing BGP path announcements. Each path in a BGP routing table is represented by an entry that consists of Network, Next Hop, Metric, LocPref, Weight and Path fields. Each BGP path with beyond right is represented with the extensions of “Total Cost”, “Link Cost”, “Transmission Rate”, “Available Time Window”, and “Expiration Time”. Total Cost: it represents currency amount for transmitting 1MB of payload through the BGP path with beyond right. It is attached to Network field (which is a destination domain address in CIDR format). Link Cost: it represents currency amount for transmitting 1MB of payload on a particular BGP link through the BGP path with beyond right. It is attached to each intermediate AS number in “Path”. If Link Cost is free on a particular link, Link Cost can be omitted. The sum of the Link Cost for all links on a BGP path sums up to Total Cost. Transmission Rate: it represents the transmission rate a BGP path with beyond right may accept. The rate of transmission is not guaranteed until a session of BGP path with beyond right is approved by all the intermediate ASs on a BGP path that perform BGP routing with beyond right. It is a new field in a BGP routing table. Available Time Window: it represents the time in a day when BGP path with beyond right may be available. It represents guessed time BGP path with beyond right is available. The time window is expressed using Greenwich Mean Time. Expiration Time: it represents the time the announced BGP path with beyond right expires. It is expected that the BGP path be re-announced by its Expiration Time, if the path should remain available. A BGP path with beyond right can be disabled by explicitly invalidating the path or by implicitly omitting re-announcement. Disseminations of BGP path beyond right is initiated by the last AS (the one close to the destination) that performs BGP routing with beyond right. In Figure 1 (b), it is AS 2 who initiates the path of 5-2-3-7. At AS 2, the BGP path (without using beyond right) of 3-7 should have been advertised by AS 3. AS 2 announces the path information to AS 5 (Link Cost is assumed to be 0.1$, other fields are omitted for simplicity): 81.16.101.0/24<0.1> 10.1.13.3 2<0.1> 3 7 Then, AS 5 announces the path information to AS 4 (Link Cost is assumed to be 0.3$): 81.16.101.0/24<0.4> 21.5.11.0 5<0.3> 2 <0.1> 3 7 If AS 5 is not willing perform BGP path with beyond right, the path announcement from AS 2 is dropped by AS 5. For the other three fields (Transmission Rate, Available Time Window, and Expiration Time), if AS 5 enforces a tighter restriction (e.g., 3MB for Transmission Rate), it updates each such field before the announcement message is forwarded to AS 4. If an existing BGP router (one that is not a BR-BGP router), it drops the BGP announcement, preventing the BGP path announcement from being propagated beyond the non BR-BGP router. It is possible for a BR-BGP router to have multiple BGP paths with beyond rights. For example, AS 2 in Figure 1 can have two BR-BGP paths (5-2-3 and 5-26-7) for network traffic from user W to Z, which consist of different BR-BGP links (Figure 3 (a)). The proposed BR-BGP routing can cope with failure of the destination-side tier-1 ISP (e.g., AS 3) as shown in Figure 3 (b). For the two BR-BGP paths in Figure 3 (a), AS 4 is the origin AS, while AS 1 is the origin AS for the BR-BGP path in Figure 3 (b). Figure 3 (c) visualizes a scenario for a failure of an internet exchange (IX) point, where the three tier-1 ISPs are interconnected (a failure of an IX point will disable interconnections of all the ISPs there). This is a scenario for a terrorist attack to a major IX point. In this scenario, although no through traffic over any tier-1 ISP is available, the interconnection of the four tier-2 ISPs will enable through traffic between W and Z. Having BGP paths using beyond rights ready in advance will minimize the impact of such attacks to major IX points. Using the feature of “traffic accounting” (described after this section), it is possible to selectively accept network traffic through BR-BGP paths in such emergency situations, which is one of the missing features in the existing BGPbased routing today. - A BGP link under beyond right 1 2 1 3 2 4 W 3 5 6 X Y 7 4 5 6 Z W X Y (a) AS-1 down 7 Z (b) AS-3 down 1 2 3 4 W 5 6 X Y 7 Z (c) IX point down Figure 3 – Examples of different scenarios for BGP routing using “beyond rights”. Signaling for path establishment and payload transmissions: Before an origin AS starts using a BGP path with beyond right, it needs to establish an active path through the BGP path by signaling each AS that performs BGP routing with beyond right (such paths are called “BR-BGP paths”) using three messages of REQ, ACK, and NACK. Each BR-BGP path may consist of multiple links, called “BR-BGP link”. Each BR-BGP link is established between two BGP routers on a BR-BGP path (they are called “BRBGP routers”). A BR-BGP path can be a part of BGP path. After a BR-BGP path is established, payload transmission starts and payload transmissions are managed using PALD, PACK, and NPACK. An existing BG-BGP path is terminated by FIN message from either a transmitting or a receiving BR-BGP router. The messages are defined and the procedure of the proposed _______ using the messages are described in the following section. REQ: it is initiated by an origin AS (e.g., AS 4 in Figure 1 (b)) and propagated to each of the BR-BGP routers (AS 5 and 2) in a path. Each REQ message consists of: Requesting AS (ASR): it is the AS number who is requesting a BR-BGP link. BGP AS list (I): the list of all the AS on the path (e.g., “5 2 3 7”) in Figure 1 (b)). Requested Transmission Rate (R): the transmission rate (in Bps) requested by the origin AS. Requested Time Duration (T): duration of the time this origin AS uses the BGP path with beyond right. Transferring AS’s nonce (X): a number randomly generated by a BR-BGP router. Transferring AS’s salt (): a number randomly generated by a BR-BGP router. Requested Message Window Size (D): the number of payload packets that carry the same transferring AS’s nonce (X), salt (), the receiving AS’s nonce (Y), and salt (). ACK: a message used by a receiving BR-BGP router to approve a request for approving BR-BGP link establishment. The message contains the following attributes: Approving AS (ASA): it is the AS number who is approving the requested BR-BGP link. BGP AS list (I): it is a copy of field “I” in the approved REQ message. Requested Transmission Rate (R): it is a copy of field “R” in the approved REQ message. Requested Time Duration (T): it is a copy of field “T” in the approved REQ message. Requesting AS’s nonce (X’): the approving AS calculates X’ = X + 1. Approving AS’s nonce (Y): a number randomly generated by a BR-BGP router. Approving AS’s salt (): a number randomly generated by a BR-BGP router. Approved Message Window Size (D): it is a copy of field “D” in the approved REQ message. Link Label (L): The approving BR-BGP router assigns a label for the approved BR-BGP link and notifies it the requesting BR-BGP router. NACK: if a receiving BR-BGP router denies a request for BR-BGP link, the receiving BR-BGP router responds to a REQ by a NACK, which consists of the following attributes: Denying AS (ASA): it is the AS number who is denies the requested BR-BGP link. BGP AS list (I): it is a copy of field “I” in the approved REQ message. Denial Factor (F): a factor that caused the denial of a BR-BGP link request (R, T, or X). Denied Factor Status: “0” if no resource is available. If some resource (less than requested by a REQ message) is available, the amount of available resource is specified (e.g., “60MBps”, while “100MBps” is requested for “R” in a REQ). After all BR-BGP links on a BR-BGP path is successfully established (i.e., all the BR-BGP routers on a BR-BGP path approves BR-BGP links), an origin BR-BGP router can initiate payload transmissions through the BR-BGP path. Each payload message contains the PALD message (it is piggy-backed to each payload packet). Receiving BR-BGP routers use the information in PALD for accounting purpose. Each receiving BR-BGP router acknowledges payload packets using PACK messages. PALD (PAyLoaD): Transmitting AS number: it is the AS number of the BR-BGP router who transmits this payload packet. It is the copy of “ASR” in the REQ for this BR-BGP link. Receiving AS number: it is the AS number of the BR-BGP router who receives this payload packet. It is the copy of “ASA” in the ACK for this BRBGP link. BGP AS list (I): it is a copy of field “I” in the approved REQ message. Link Label (L): it is the copy of “L” in the ACK for this BR-BGP link. Transferring AS’s nonce (Xi,j): the j-th transferring AS’s nonce in the i-th round of the BR-BGP link (described in the next section). Receiving AS’s nonce (Yi,j): the j-th transferring AS’s nonce in the i-th round of the BR-BGP link (described in the next section). Size of the payload (S): it is the size of the payload contents (the size of a network-layer packet, including the network-layer packet header) in bytes. PACK (Payload ACKnowledge): Transmitting AS number, Receiving AS number, BGP AS list (I), Link Label (L), and Size of the payload (S): same as they appear in a PALD message. Receiving AS’s nonce (Yi,j): the j-th transferring AS’s nonce in the i-th round of the BR-BGP link (described in the next section). An existing BR-BGP link can be terminated either a transmitting or a receiving BR-BGP router, by releasing a FIN message, which consists of the following attributes: FIN (FINish): Transmitting AS number, Receiving AS number, BGP AS list (I), and Link Label (L): same as they appear in a REQ message. Transferring AS’s nonce (Xi,j): the j-th transferring AS’s nonce in the i-th round of the BR-BGP link (described in the next section). Receiving AS’s nonce (Yi,j): the j-th transferring AS’s nonce in the i-th round of the BR-BGP link (described in the next section). The procedure of the BR-BGP path setup, payload transmissions through an existing BR-BGP path, and a termination of an existing BR-BGP path is visualized in Figure 2, while the one for the detailed payload transmissions is shown in Figure 3. In Figure 2, a BR-BGP path setup is initiated by an origin BR-BGP router (a BR-BGP router in AS 4 for a BR-BGP path of 4-5-2-3 in Figure 1 (b)). The BRBGP router in AS 4 first constructs its REQ message, and it sends the REQ to the BR-BGP router in AS 5. If the router in AS 5 rejects the request, it constructs a NACK message and sends the NACK to the BRBGP router in AS4. Otherwise, it creates an account for this BR-BGP path, creates a valid BR-BGP link label (L) and constructs a new REQ message and sends the REQ message to the BR-BGP router in AS 2 (the BR-BGP router in AS 5 holds its ACK to the one for AS 4 until it receives an ACK from the one in AS 2. This procedure repeats until a REQ message reaches the BR-BGP router in AS 3. If the BR-BGP router in AS 3 approves the request for its BR-BGP link, it causes the router to generate an ACK message, or a NACK message will be released by the router. All the intermediate BR-BGP routers send either an ACK or a NACK all the way back to the origin BRBGP router (i,.e., the one in AS 4 in Figure 1 (b)). AS4 AS2 AS3 AS7 Time Payload Transmissions AS5 Figure 2 - the timeline view of the BR-BGP routing signaling and payload-transmission procedure Traffic accounting: The proposed ____ relies on tier-2 and 3 ISPs who are willing to share the link bandwidth of their peer links when they are underutilized at the same time the transmission bandwidth they accepted may cause increase in their payment to their providers (i.e., tier-1 ISPs for tier-2 or tier-2 ISPs for tier-3 ISPs). Regarding peer-to-peer links, the freeride of their transmission bandwidth by other ISPs (other than those from the two ISPs who are the two ISPs of the edges of a peer-peer link) will discourage having such peer-peer links from the beginning. To facilitate sharing their transmission bandwidth with peers, we also propose a mechanism of monetary incentive, which realizes an ecosystem with network transmission bandwidth and payments as the two primary factors of the system. When payload transmissions on a BR-BGP path start, the amount of the payload traffic must be confirmed by both transmitting and receiving BRBGP routers on a BR-BGP link. For secure payload traffic accounting, there are two requirements in nonrepudiation, which are defined as follows: (a) Sender repudiation: when a sender sends a message to a destination and the message is successfully delivered to the destination, the sender can not deny to have sent that message to the destination. This means that if a sender has transmitted a certain amount of traffic to a BRBGP router through an already-established BRBGP path (and the traffic successfully reached the router), the sender can not deny the transmission of the message (thus the BR-BGP router is sure for the payment for that network traffic). (b) Sender masquerading by replays: free-riding on an existing BR-BGP path using replays by a malicious third party must be prevented. Two BR-BGP routers on an active BR-BGP path must uniquely identify each payload packet the legitimate BR-BGP router transmits for accurate charging, which can be performed using key attributes such as secret BR-BGP path ID and packet sequence number. However, encrypting these attributes will not prevent such attacks, since malicious third parties may intercept and release the copies of them later (the patterns of sequence number may be identified) for hijacking existing BR-BGP paths. The following procedure is used to eliminate the two risks. When the first BR-BGP router (called “the origin BR-BGP router, which is a BR-BGP router in AS 4 in Figure 1(b)) constructs is REQ message, it randomly generates a number (“Transferring AS’s nonce (X)”), at least in a way it is hard for anyone to predict the number. It becomes “X1,1”, which is the first X in the first __ of ____, as well as another random number (“Transferring AS’s salt ()”). The first “” becomes “1”, which is the one for ______. The parameter “D1” represents the number of a sequence of “Xi,j” that are generated from “X1,1”. Then, the whole REQ message is encrypted by the public key of a receiving BR-BGP router (the one in AS 5 in Figure 1(b)). When the receiving BR-BGP router received the REQ message, it decrypts the message using its private key. The router generates its pairs of Approving AS’s nonce (Y) and Approving AS’s salt () in the same way the transmitting BR-BGP did for its X and . The first “Y” generated by the receiving BR-BGP routers becomes “Y1,1” and the first “” becomes “1”. If the receiving BR-BGP router agrees with “D1” in the REQ, it simply copies its value to “D’1” to its ACK. The router adds constant value 1 to X1.1 in the REQ. The receiving BR-BGP router constructs an ACK message and encrypts the whole message using the public key of the transferring (requesting) BR-BGP router. See Figure 3 for this. - A parameter generated by a transmitting BR-BGP router - A parameter generated by a receiving BR-BGP router Sender X1,1 1 Receiver D1 Y1,1 1 D1’ X1,2 = X1,1 + 1 X1,3 = X1,2 + 1 Y1,2 = X1,2 + 1 X1,D-1 = X1,D-2 + 1 Time X1,D = X1,D-1 + 1 Y1,3 = X1,3 + 1 Y1,D-1 = X1,D-1 + 1 X1,D = X1,D + 1 Figure 3 – the BR-BGP procedure for processing each payload packet When the transferring BR-BGP router receives an ACK from the receiving BR-BGP router, it constructs a PALD message in the following way. The router calculates X1,2 by X1,2 = X1,1 + . The value of Y is updated by Y1,1 + 1. Then, the whole PALD message is encrypted by the public key of the receiving BR-BGP router. The transferring BR-BGP router then transmits the PALD message to the receiving BR-BGP router as it is piggy-backed to the first payload message. When the receiving BR-BGP router receives this PALD message, it performs the procedure in Figure 4. Using i, the receiving BR-BGP router can calculates Xi,j from Xi,(j-1) (since Xi,j = Xi,(j-1) + i) at the same time the receiving BR-BGP router uses “Yi,(j-1) + 1” in the PLAD message to verify if this PLAD is from the transmitting BR-BGP router (but not from anyone else) since Yi,j in the ACK message is encrypted by the public key of the transmitting BR-BGP router. Only the right transferring BR-BGP router can correctly decrypt Yi,j in the ACK (as a response to a REQ) and the transferring BR-BGP encrypts Yi,j using the receiving BR-BGP router’s public key after Yi,j is increased by the transferring BR-BGP router. This eliminates the possibility of replay attacks, by someone who masquerades to be the transferring BRBGP router. The confirmed “Yi.j + 1” (confirmed at the receiving BR-BGP router) guarantees no repudiation by the transferring BR-BGP router. 1. Decrypt the PALD message using its private key. 2. Extract (Xi,j) and (Yi,j-1 + 1) from the PALD message and compare them with their local copies. If they match, proceed to 3. Otherwise simply drop the whole payload message. 3. Compare the actual payload size and the declared payload size (i.e., the value of “S” attribute in the PALD message). If they match, proceed to (4). Otherwise send a NPACK to the transmitting BR-BGP router. 4. Update the account information (e.g., the total amount of traffic load that went through the BR-BGP link for charging purpose). 5. Construct a PACK message in the following way. First, the router calculates Yi,j by Yi,j = Xi,j + i (e.g., Y1,2 = X1,2 + 1). 6. Encrypt all the attributes in the PACK using the public key of the transmitting BR-BGP router. 7. Send the PACK message to the transmitting BR-BGP router. Figure 4 When the transferring BR-BGP router receives a PACK message from the receiving BR-BGP router, it performs the procedure in Figure 5. Using i, the transferring BR-BGP outer calculates “Y’i,(j+1)” as Y’i,(j+1) = Yi,j + i. It decrypts the PACK message using its private key, extracts Yi,(j+1), and compares Yi,(j+1) and Y’i,(j+1). This procedure makes sure that the PACK is from the correct receiving BR-BGP router. The values of and should be changed once in a while to prevent any third party from estimating their values to eliminate a chance of replays. To achieve this goal, two additional message types of RENEW and RACK are used. Before the transferring BRBGP router transmits as many as Di payload messages, it is expected to issue a RENEW message to the receiving BR-BGP router. A RENEW message is inserted between two payload messages from a transferring to a receiving BR-BGP routers and the message is authenticated by (Xi,j) and (Yi,j + k), where k is the k-th message from the beginning of a new and (see Figure 3). The new values of and should be used after as many as Di payload messages have been transmitted by the transferring BR-BGP router. An existing BR-BGP link can be terminated by either a transferring or a receiving BR-BGP router using a FIN message. Each FIN message is authenticated using Xi,j and Yi,j in the same way each of PALD message is authenticated. If an intermediate receiving BR-BGP router initiates a FIN message, a chain of FIN messages are propagated in both directions (i.e., upstream and downstream of the payload flow), which terminates an entire BR-BGP path. 1. Decrypt the received PACK message using its own private key. 2. Extract Yi,(j+1) from the PACK. 3. Locally calculate Y’i,(j+1) as Y’i,(j+1) = Yi,j + i. 4. Compare Yi,(j+1) in PACKi,j and Y’i,(j+1). If they do no match, proceed to a process of error recovery. Otherwise terminate this procedure (success). We propose a three-layer mechanism to achieve universal, but less financially efficient, electric payment system for the proposed ________. The three layers are (1) payer-payee specific virtual payment, (2) universal virtual payment, and (3) payment by real currency (Figure 6) using multiple layers of payment brokers. A payment broker is a process that manages payments for BR-BGP traffic. A group of ASs that often perform BR-BGP transmissions become the members of a payment broker, which becomes the lowest level payment broker (Figure 7). The lowest level payment brokers are the one who manage payment for BR-BGP traffic for lowest tier ASs. Mutual Payment Mutual Payment Payer-payee Specific Virtual Payment Universal Virtual Payment Figure 5 – Payment model Through network traffic with beyond right will not directly contribute to the own business of an ISP that accepts such through network traffic. Such through traffic can even negatively impact to an ISP’s own business since handling through traffic consumes networking resources. It is harmful to own business even when ISPs have unutilized networking resources, such as transmission bandwidth, since sharing unutilized link bandwidth requires their prior announcements to other ISPs. Once link bandwidth is announced to other ISPs, an announcing ISP may receive through network traffic or BR-BGP path setup requests when it has no unutilized transmission bandwidth on a peer link. To encourage sharing link bandwidth and publishing AS-peer connections to others despite these possible disadvantages, some incentive is necessary for encourage BR-BGP routing with beyond right. Any form of electric payment that is good only for a specific pair of two ISPs is not flexible enough. This form of electric payment requires a payee to convert the electric payment made by a payer. The electric payment made by a specific payer to a specific payee can not be used for payment to any other payee. The electric payment method proposed by ______ [0, 0, 0] belongs to this category. If a receiving ISP wants to use the credit for payment to other ISPs (i.e., if the receiving ISP becomes a transferring ISP to others), this type of electric payment is inefficient since each conversion of virtual currency to real currency causes significant conversion fee. This conversion fee can be a major factor that discourage adoption of BR-BGP routing with beyond rights. Convert Convert Payment by Real Currency Figure 6 – the three-layer payment model AS connection tree Payment broker tree Tier i AS Tier (i-1) AS Tier (i-2) AS Tier (i-3) AS Figure 7 – Payment broker layers for different levels of ASs When an origin BR-BGP router transmits network traffic through a BR-BGP path, intermediate BRBGP routers charge transmission fee based on the traffic load. Intermediate BR-BGP routers represent their ISPs, and charge transmission fee to an origin BR-BGP ISP on behalf of for their home ISPs. The ISP of the origin BR-BGP router makes payment to each such of ISPs as payer-payee specific virtual payment. The virtual currency in payer-payee specific virtual payment should be used for mutual payment between the two ISPs (i.e., if the receiving ISP becomes the transferring ISP for the first transferring ISP), which eliminates a need for converting the virtual currency in payer-payee specific virtual payment to the one for the universal virtual payment. If receiving ISPs do not have a chance to use the virtual currency in payer-payee specific virtual payment for transferring through traffic through a transferring ISP, each of them can convert the virtual currency in payer-payee specific virtual payment to the one in the universal virtual payment. Using the universal virtual currency, the credit for handling a particular transferring ISP’s through traffic with beyond right can be used as a mean for payment to other ISPs, when this receiving ISP becomes a transferring ISP to other ISPs in the Internet. This allow ISPs who perform BR-BGP routing to minimize their needs to convert their universal virtual currency to real currency, which causes significant amount of conversion fee. Payment brokers in the lowest level form a group and they become members of a payment broker in the second lowest level, while payment brokers in the second lowest level become members of a payment broker in the upper level. A top-level payment broker is for a tier-1 AS, who manages payments for any BR-BGP traffic generated by any AS under its provider-customer relations. Finally, those tier-1 payment brokers form a mesh interconnections to resolve payments between two ASs that belong to different tier-1 ASs (Figure 8). F R105 B R11 R1 R3 R102 E C R2 R12 R104 A R101 R14 R103 D R13 - AS Universal Virtual Payment - Payment Broker Figure 8 – In Figure 8, AS A and B belong to a lowest-level payment broker, R101. When A makes a payment to B, the payment is made by a payee-payer specific payment using R101’s broker-local virtual payment. After B receives A’s payment, B can use the payment from A for its payment to other AS (e.g., AS C), first using R101’s broker-local virtual payment, which is exchanged to R11 and R102’s broker-local virtual payment. When C makes a payment to another AS, D, the payment will be made using the same procedure, but using the broker-local virtual payment of R2. When an AS makes payment to other ASs who belong to other tier-1 ASs (i.e., those who belong to other top-level payment brokers), the payment is first exchanged to the universal virtual currency. For example, when D makes a payment to E, who belongs to another top-level payment broker (R3), the payment by D is first exchanged to the universal currency at R2 before it is transferred to R3. When R3 receives the payment by D in the universal virtual currency, the payment is exchanged to the brokerlocal virtual payment of R3, R13, and R104 in that before E receives the payment from D. Payments by E to other ASs who belong to different tier-1 ASs (e.g., F) can be performed in the same way, using the credit (payment) that originated from A. Why this mechanism? The proposed ____ provides a mechanism to flexibly resolve payment arrangements from a paying AS to a payee AS in two different ways: direct payments and proxy payments. Consider a following case. AS 4 is the origin AS of a BR-BGP transmission using a BR-BGP path, 4-5-2-3-7, in which 5-2 and 2-3 are BR-BGP links (Figure 9 (a)). Direct payment: an origin AS makes payment to the intermediate ASs that have carried the origin AS’s BR-BGP traffic (Figure 9 (b)). Then, the origin AS (i.e., AS 4) makes a request for imbursement to its provider AS (i.e., AS 1). If the requested amount of the reimbursement is less than the fee the provider AS charges to AS 4, AS 1 can reimburse the payment by subtracting the connection fee 4 usually pays to 1. Direct payment is performed using the following procedure. Direct payments are performed using two messages of Payment Request (PR) and Payment Confirmation (PC) messages. First, AS 5 and 2 send a PR (PR5 and PR2 respectively) to 4 for requesting payment. AS 4 responds to each of the PR messages by a PC message, confirming its willingness for the payment. Actual payment will be performed by AS 4 after it issues the PC messages. Proxy payment: an origin AS makes a proxy payment request to an AS that is responsible for the cause of a BR-BGP traffic (Figure 9 (c)). For example, when AS 5 requests a payment to 4 for 4’s BR-BGP traffic through it, AS 4 requests AS 1 to make a payment to AS 5 for AS 4. If AS 4 requests another AS, typically its provider (e.g., AS 1), to make a payment for 4 (assuming that it is AS 1’s failure that necessitates AS 4 to generate the BR-BGP transmission and that there is a prior agreement for the payment between AS 1 and 4), on arrival of a PR message (e.g., PR5), AS 4 forwards the PR message to AS 1. AS 1 then confirms the payment to AS 5 on behalf of AS 4 by sending a PC message to AS 5 (e.g., PC5). (a) a BR-BGP transmission through a BR-BGP path 4-5-2-3-7 4 2 5 BR-BGP link BGP peering link 7 3 - Origin AS - 1 PC2 PR5 PC5 4 2 5 Destination or intermediate AS 3 PR5 PR2 (b) Direct payment by an origin AS (AS 4) 7 PC5 PC2 PR2 4 2 5 3 PR5 PR2 (c) Proxy payment by a provider AS (AS 1) 7 Figure 9 – Two payment methods of direct and proxy payments 4. Performance evaluation (“assessment”?) (a) Theoretical analysis (b) Prototyping 5. Conclusions and Future Work Analysis of the current domain interconnections in the Internet to assess the effectiveness of the proposed SDN-based network traffic control. Using simulation, the effectiveness of the proposed SDN-based routing will be studied. The overhead of performing the proposed SDNbased routing will be studied. References [a] Lixin Gao, “On Inferring Autonomous System Relationships in the Internet,” IEEE/ACM Transactions on Networking, vol. 9, no. 6, pp. 733-745, December 2001. [b] Hiroshi Fujinoki and Andrew Hauck, “Analysis on the Current and the Future Internet Structure Regarding Multi-homed and Multi-path Routing,” Journal of Internet Services and Applications, vol. 2, no. 3, pp. 257-270, November 2011. [0] RFC 5151 Inter-Domain MPLS & GMPLS RSVP-TE Extensions February 2008.