Network Working Group Muneyoshi Suzuki INTERNET DRAFT NTT Expires May 25, 1997 November 25, 1996 Architecture of the Resource Reservation Service for the Commercial Internet Status of this Memo This document is an Internet-Draft. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress". To learn the current status of any Internet-Draft, please check the "1id-abstracts.txt" listing contained in the Internet-Drafts Shadow Directories on ftp.is.co.za (Africa), nic.nordu.net (Europe), munnari.oz.au (Pacific Rim), ds.internic.net (US East Coast), or ftp.isi.edu (US West Coast). Abstract The purpose of this document is to clarify the architecture that should be used for the resource reservation service for the commercial Internet. First, this document explains the basis of the tariff for current telecommunication and Internet services. Then it clarifies problems in the billing for Internet service, and describes how billing can be improved by using the resource reservation setup protocol. This document also studies technical and application models for a commercial resource reservation service model, and clarifies an architecture for the resource reservation setup protocol. Last, it explains why ATM is an indispensable implementation technology for the commercial resource reservation service, and investigates an implementation method for a resource reservation setup protocol that is based on ATM. Suzuki Expires May, 1997 [Page 1] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 1. Introduction With the development of new multimedia applications, such as voice, audio, picture, and video communication, the demands on the resource reservation service are increasing, especially as Internet traffic volume grows explosively due to these applications. Therefore, tariff systems for Internet service have tended to adopt measured rate billing, and the resource reservation setup protocol [2, 3] is increasingly important as a method for implementing measured rate billing. The resource reservation setup protocol [1] must support billing if it is to be applied to the commercial Internet, especially measured rate billing between interconnected Internet Service Providers (ISPs) is needed. And to investigate the implementation method of the resource reservation setup protocol in the commercial Internet environment is also needed. The purpose of this document is to clarify the architecture that should be used for the resource reservation service for the commercial Internet. First, this document explains the basis of the tariff for current telecommunication and Internet services. Then it clarifies problems in the billing for Internet service, and describes how billing can be improved by using the resource reservation setup protocol. This document also studies technical and application models for a commercial resource reservation service model, and clarifies an architecture for the resource reservation setup protocol. Last, it explains why ATM is an indispensable implementation technology for the commercial resource reservation service, and investigates an implementation method for a resource reservation setup protocol that is based on ATM. Incidentally, it is essential to examine billing based on business administration issues, not technical ones. For example, on a telephone service, it technically makes sense to charge the caller when the user being called is on another line. This is because, telephone switches were in operation when they notified the caller that the number he called was busy. However, such a billing policy is contrary to the customs of business. Readers should note that the billing problems and solutions discussed in this document are not only based on the technical viewpoint. Suzuki Expires May, 1997 [Page 2] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 1.1 The Basis of the Tariff Basic elements that determine the network tariff are distance, bandwidth, time, and information volume. In many cases, the network tariff reflects the link cost to some extent. Note: In this document, distance means the distance between the regions where users call from and to, not the actual length of the physical links that connect users. In actual communication, a route depends on network situations, so a charge based on the physical link distance is inappropriate. 1.2 The Tariff in Telecommunication Services Classifications of the basic styles of tariff systems in telecommunication services and some examples are shown below. The following classifications do not cover applied billing styles, for example contents-based charging or premium charging such as the Dial Q2 service of NTT, or the 900 telephone service. o Flat-rate billing - Leased line In most cases, the tariff is based on distance and bandwidth. - PVC-based frame relay and ATM In most cases, the tariff is based on distance and bandwidth. o Measured-rate billing - Telephone In most cases, the tariff is based on distance and time. - Circuit switching In most cases, the tariff is based on distance, bandwidth, and time. - SVC-based frame relay and ATM In most cases, the tariff is based on distance, bandwidth, and time, or information volume. - X.25 packet switching In most cases, the tariff is based on information volume. Furthermore, measured rate billing is classified into calling- or called-party billing. The basic charge style for telecommunication service is calling-party billing. Suzuki Expires May, 1997 [Page 3] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 o Calling-party billing - Usual telephone service o Called-party billing - The Free Dial service of NTT and the 800 telephone service. Basically, telecommunication service is designed for connection- oriented, point-to-point, and bidirectional communication. In the case of measured-rate billing, usually the calling or the called party pays the bidirectional communication charge. In the case of called-party billing, a function that allows incoming calls to be accepted or refused based on the calling-party address, or a function that restricts the calling-party addresses that are permitted to use called-party billing, is indispensable. This is because, the communication charge that a called party is willing to accept is usually limited. The current tendency in telecommunication-service tariff systems is to change from measured-rate billing, which reflects link costs accurately, to flat-rate billing, which simplifies the charging system, and service tends to be provided by flat-rate billing inside a single provider. The tariff for services between provider is usually the sum of the individual providers charges. Flat-rate billing, like that within a single provider, is not currently realistic for services that cross providers. 1.3 Tariffs for Internet Service Classifications of basic styles and examples of the tariff system for Internet service are shown below. o Flat-rate billing - Internet access via leased line, PVC-based frame relay, or ATM In most cases, the tariff is based on bandwidth. o Measured-rate billing - Dialup Internet access using a modem or N-ISDN In most cases, the tariff is based on time. - Internet access via leased line, PVC-based frame relay, or ATM Some ISPs have adopted information-volume-based tariff systems. Suzuki Expires May, 1997 [Page 4] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 Note: Dialup access charges in this document do not include the basic telephone fee. Until now, the tariff system for Internet access has mainly been flat-rate billing, because measured-rate billing is technically difficult to implement. However, the development of new multimedia applications, such as voice, audio, picture, and video communication, has caused the traffic over the Internet to increase explosively. The cost of using the public network service is lower than when using a private network system, if users can share equipment and lines. However, if the traffic from all its users is at a steady high rate, the cost advantage of the public network service is lost. Therefore, Internet service tariff systems, although they use leased line access, tend to adopt information-volume-based tariff systems. 2. Billing in the Resource Reservation Service 2.1 Problems of Billing in Internet Service Basically, the tariff system for Internet service seems similar to that for telecommunication service. However, note that the tariff system for Internet service based on the access method from the user site to the ISP, is not based on the end-to-end communication method. The Internet is a connection less and unreliable communication, and some users are beginning to use it for multicast communication. But, the telecommunication is basically a connection-oriented, point-to- point, and bidirectional form of communication, so telecommunication and Intrenet communication are essentially different in some ways. Current Internet service does not allow billing based on end-to-end user site distance. This is because the structure of the IP address is flat, rather than a layered structure that contains information about the provider and region. So information about distance for billing purpose cannot be obtained from the IP address directly. Note: In this document, an address means an identifier, such as a class A, B, or C IP address, that uniquely distinguishes an end point. It does not means a group identifier such as a class D address. For Internet service, billing based on bandwidth can be provided, but only for the line bandwidth between the user site and the ISP; it is not based on the end-to-end user or application bandwidth, such as the bandwidth in telecommunication services. Suzuki Expires May, 1997 [Page 5] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 Current Internet service, except for the dialup access, does not allow billing based on time because the IP is a connection less communication, and timing information about the beginning and ending of billing is too difficult to obtain. Some current ISPs have adopted information-volume-based tariff systems. However, this billing is based on the information volume of IP packets that are sent to or received from the user site and the ISP. Again, because the IP is a connection less and unreliable communication, it is too difficult to provide billing based on the information volume of IP packets that are actually used between end- to-end users or applications. It is not impossible to provide both user billing, and application provider billing over the Internet when particular services are used. These forms of billing are equivalent to calling- and called-party billing in telecommunication service. However, obtaining the timing information about the beginning and ending of application usage at the IP layer is too difficult because the IP is a connection less communication. To have billing based on usage time, the service application responsible for the bill must identify the user and monitor the usage. Also a billing process, where part of the billing is transferred from the user to the service provider, must be implemented. As a result, the billing system complexity is increased. 2.2 Improved Billing Using the Resource Reservation Setup Protocol As explained above, billing for current commercial Internet service has many problems, but a resource reservation setup protocol may solve these problems. For example, the resource reservation setup protocol ensures the availability of end-to-end network resources, so billing based on bandwidth (FlowSpec) between user sites may be possible. Also, the resource reservation setup protocol explicitly shows the resource acquire and release timings, so billing based on time may be feasible. The resource reservation setup protocol also guarantees QoS based on the FlowSpec, so billing based on the information volume that is actually used between end-to-end users or applications may also be feasible. Furthermore, there is a possibility that the billing for each IP flow can be distributed to ether the sender or the receiver. Suzuki Expires May, 1997 [Page 6] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 However, the resource reservation setup protocol cannot solve the problem of how billing can be based on distance because the flat structure of the IP address does not change and it is still impossible to obtain information about distance for billing from the IP address directly even when the resource reservation setup protocol is used. 3. Model of a Commercial Resource Reservation Service 3.1 Technical Model of the Resource Reservation Service This subsection looks at unreliable, unidirectional, and tree structured multicast architecture as a technical model for a resource reservation service. For the time being, the QoS to all receivers is assumed to be the same. The following section examines heterogeneous QoS and filter spec. +---+ | S | +---+ +-----------------/---\-----------------+ | a / \ b | | / \ | | / ISP-A /\ | | / / \ | | / c / \ d | +-----------/--------/------\-----------+ +---+ +-------+ +---+ |R1 | |router | |R2 | +---+ +-------+ +---+ +-----------------/---\-----------------+ | e / \ f | | / \ | | / ISP-B /\ | | / / \ | | / g / \ h | +-----------/--------/------\-----------+ +---+ +---+ +---+ |R3 | |R4 | |R5 | +---+ +---+ +---+ Fig. 3.1: Resource Reservation Service Model. As shown in Fig. 3.1, ISP-A and ISP-B are interconnected, a sender S and receivers R1 and R2 belong to ISP-A, and receivers R3, R4, and R5 belong to ISP-B. This subsection studies the receiver billing and the sender billing resource reservation service with this model. Suzuki Expires May, 1997 [Page 7] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 3.1.1 Receiver billing When the resource reservation service is provided under receiver billing, the problem is how to bill for the shared links, such as b, c, and f. The shared link cost must be distributed and billed to receivers based on some rule. One solution inside a single ISP is to adopt a tariff system that does not depend on how the links shared between receivers. Billing that is based on the cost of the links that make up the multicast tree is equivalent to billing based on distance. Therefore, billing that does not depend on the link sharing approach is equivalent to billing that is not based on distance. This means the billing can be based on bandwidth (FlowSpec), time, and information volume. For example, if an interconnected destination ISP is regarded as a receiver, ISP-A bills to R1, R2 and ISP-B, and ISP-B bills to R3, R4, and R5 [1]. The billing from ISP-A to ISP-B is distributed based on some rule, and is added to the base charge in ISP-B. If a large number of users join the multicast and the statistical tendency of network utilization is known, it is possible to provide this type of tariff system, although it does not accurately reflect communication costs. Another solution is to distribute the shared link cost among the receivers that share the link. For example, the cost of link b would be shared by R2, R3, R4 and R5. This method does reflect accurate communication costs. However, in practice it is difficult to implement the billing system since the complexity of computing the cost of the shared link, located near a sender like b, is increased because the receiver can dynamically join and leave the multicast tree. Therefore, in the case of receiver billing, if many users join the multicast and the statistical tendency of network utilization is known, billing based on bandwidth (FlowSpec), time, and information volume can be provided. 3.1.2 Sender billing When the resource reservation service is provided under the sender billing, the problem due to the shared link is avoided, because there is no need to distribute the shared link cost. In the above model, the sender would be billed for the link costs from a to h. Suzuki Expires May, 1997 [Page 8] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 Therefore, with sender billing, billing based on accurate link costs can be provided. Billing based on the link cost is equivalent to billing based on distance. However, information about distance for billing cannot be obtained from the IP address directly. Therefore, a database that can extract information about distance from the destination IP address is needed to enable billing based on the link cost. This is also true for receiver billing: if a number of users join the multicast and the statistical tendency of network utilization is known, it is possible to provide billing based on bandwidth (FlowSpec), time, and information volume. That is, the sender pays for the billing to R1, R2, and ISP-B in ISP-A, and to R3, R4, and R5 in ISP-B. Therefore, with sender billing, if a database is implemented that can extract information about distance for billing from the destination IP address, it will be possible to provide billing based on distance, bandwidth (FlowSpec), time, and information volume. And if many users join the multicast and the statistical tendency of network utilization is known, it will also be possible to provide billing based on bandwidth (FlowSpec), time, and information volume. 3.2 Application Model for the Resource Reservation Service This subsection examines the following multimedia applications to develop an application model for the resource reservation service. o Broadcast-type application model o Advertisement-type application model o Conference-type application model Methods of implementing the application model using the technical model described in the previous subsection are also examined. 3.2.1 Broadcast-type application model We assume that the broadcast-type application model has the following features. o The application provider broadcasts to receivers using the multicast, and, in practice, the application is open to the public. Suzuki Expires May, 1997 [Page 9] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 o Many receivers subscribe to the broadcast, and the statistical tendency of network utilization is known. o Joining the multicast tree is initiated from the receiver. o The receiver pays the full amount billed. o The billing is based on information volume or bandwidth (FlowSpec) and time, and not on distance. Features of this model correspond to receiver billing in the technical model, so it is appropriate for this model to be supported by it. Therefore, receiver billing based on bandwidth (FlowSpec) and time, or information volume can be provided. 3.2.2 Advertisement-type application model We assume that the advertisement-type application model has the following features. o The application provider advertises to receivers using the multicast, and, in practice, the application is open to the public. o Many receivers subscribe to the advertisement, and the statistical tendency of network utilization is known. o Joining the multicast tree is initiated from the receiver side. o The application provider pays the full amount billed. o A function that restricts the region in which the receiver is permitted to join, or a function that decides whether to accept or refuse the joining request based on the IP address of the receiver or based on the tariff to be billed, is indispensable. This is because the communication charge that is acceptable to an application provider is usually limited. Features of this model roughly correspond to sender billing in the technical model, so it is appropriate for this model to be supported by it. But this model needs a function that restricts the region in which the receiver is permitted to join, or a function that decides whether to accept or refuse the joining request based on the IP address of the receiver or based on the tariff to be billed. If the region that the receiver is permitted to join is simply restricted by the ISP boundary, the model can be implemented by restricting the IP flow forwarding between ISPs. Suzuki Expires May, 1997 [Page 10] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 But if the decision to accept or refuse the joining request is based on the IP address of the receiver or based on the tariff to be billed, a database that can extract information about permission or distance for billing from the destination IP address is needed. In the resource reservation setup protocol, a procedure that supports this kind of processing is also needed. However, if this procedure is processed only by the sender, and the number of receivers significantly increases, saturation of the sender protocol processing may occur. Therefore, an intermediate node is needed inside the multicast tree, and this intermediate node will decide whether to accept or refuse the joining request. Therefore, in the advertisement-type application model, if the region that the receiver is permitted to join is simply restricted by the ISP boundary, it is appropriate for this model to be supported by the sender billing in the technical model. Thus, sender billing based on bandwidth (FlowSpec) and time, or information volume, can be provided. If the decision to accept or refuse the joining request is based on the IP address of the receiver or based on the tariff to be billed, in addition to the sender billing in the technical model, a database, that can extract information about permission or distance for billing from the destination IP address is needed. In the resource reservation setup protocol, a procedure that supports this process is also needed. In this case, it can be provided by sender billing based on distance, bandwidth (FlowSpec) and time, or information volume. 3.2.3 Conference-type application model. We assume that the conference-type application model has the following features. o The conference is held by a small number of participants. o The statistical tendency of network utilization in the conference depends on each conference style and the tendency is hard to estimate. Suzuki Expires May, 1997 [Page 11] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 o Joining the conference is initiated by each participant. That is - Joining the multicast tree from an existing participant or receiving information from the conference server is initiated by the receiver. - Construction of the multicast tree for existing participants or information sent to the conference server is initiated by the sender. o Management of conference participants is indispensable. A function that can decide to accept or refuse a participation request based on the IP address of the potential participant, or a similar function is needed. To avoid establishing an unreasonably expensive tariff for short distance communications, this model needs billing based on accurate link costs, because the tendency of network utilization is hard to estimate. Therefore, in this model, in addition to the sender billing from the technical model, a database that can extract information about permission and distance for billing from the destination IP address is needed. In the resource reservation setup protocol, a procedure that supports this process is also needed. In this case, it can be provided by sender billing based on distance, bandwidth (FlowSpec) and time, or information volume. 3.3 Architecture of the Resource Reservation Setup Protocol A combination of the billing side and the reserve initiating side in the resource reservation setup protocol based on the above studies is shown in Table 3.1. Table 3.1: Combination of Billing and Reserve Initiating Sides. +---------------+---------------+---------------+ | Application | Billing Side |Initiating Side| +===============+===============+===============+ | Broadcast | Receiver | Receiver | +---------------+---------------+---------------+ | Advertisement | Sender | Receiver | +---------------+---------------+---------------+ | Conference | Sender |Sender,Receiver| +---------------+---------------+---------------+ Suzuki Expires May, 1997 [Page 12] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 In addition to supporting all combinations of the billing side and the reserve initiating side shown above, the commercial resource reservation service must satisfy the following requirements for sender billing. o A function is needed that restricts the region that a receiver is permitted to join, or that decides whether to accept or refuse the joining request based on the receiver IP address and/or on the tariff to be billed. o If the application is open to the public, an intermediate node that decides whether to accept or refuse the joining request is needed inside the multicast tree. A resource reservation setup protocol that satisfies these requirements can be achieved with the following sender initiation architecture. Suzuki Expires May, 1997 [Page 13] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 o The basis of the resource reservation setup protocol is sender initiation. That is, as shown in Fig. 3.2, the sender explicitly designates the receiver address, sends a resource reservation setup message (SETUP), and constructs the multicast tree. +---+ +--------| S |--------+ | +---+ | +--------|--------/-|-\--------|--------+ | | / | \ | | | | / | \ | | | SETUP / SETUP /\ SETUP | | | / | / \ | | | | / | / \ | | +--------|--/------ | ------\--|--------+ +-V-+ +---V---+ +-V-+ |R1 | | | |R2 | +---+ |router | +---+ +------| |------+ | +-------+ | +--------|--------/-|-\--------|--------+ | | / | \ | | | | / | \ | | | SETUP / SETUP /\ SETUP | | | / | / \ | | | | / | / \ | | +--------|--/------ | ------\--|--------+ +-V-+ +-V-+ +-V-+ |R3 | |R4 | |R5 | +---+ +---+ +---+ Fig. 3.2: Sender Initiation. Suzuki Expires May, 1997 [Page 14] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 o In the case of receiver initiation, as shown in Fig. 3.3, the receiver explicitly sends the joining request message (JOIN), and if the sender accepts it, the sender sends a resource reservation setup message to the receiver. +---+ | S | +A--+ +-----------------/|-|\-----------------+ | / | | \ | | / | | \ | | JOIN| |SETUP | | / | | / \ | | / | |/ \ | +-----------/------|-|------\-----------+ +---+ +----V--+ +---+ |R1 | | | |R2 | +---+ |router | +---+ +-------> | | +---- +-------+ +-------|-|-------/---\-----------------+ | | | / \ | | | | / \ | | JOIN| |SETUP /\ | | | | / / \ | | | | / / \ | +-------|-|-/--------/------\-----------+ +--V+ +---+ +---+ |R3 | |R4 | |R5 | +---+ +---+ +---+ Fig. 3.3: Receiver Initiation. Suzuki Expires May, 1997 [Page 15] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 o However, if the application is open to the public, as shown in Fig. 3.4, the intermediate node inside the multicast tree that decides whether to accept or refuse the joining request, may send a resource reservation setup message as a response to the joining request message. +---+ | S | +---+ +-----------------/---\-----------------+ | / \ | | / \ | | / /\ | | / / \ | | / / \ | +-----------/--------/------\-----------+ +---+ +-------+ +---+ |R1 | | | |R2 | +---+ | node | +---+ +-------> | | +---- +-------+ +-------|-|-------/---\-----------------+ | | | / \ | | | | / \ | | JOIN| |SETUP /\ | | | | / / \ | | | | / / \ | +-------|-|-/--------/------\-----------+ +--V+ +---+ +---+ |R3 | |R4 | |R5 | +---+ +---+ +---+ Fig. 3.4: Intermediate Node. Suzuki Expires May, 1997 [Page 16] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 4. Implementation Method of the Resource Reservation Setup Protocol 4.1 Datalink Media for the Commercial Resource Reservation Service The resource reservation setup protocol can designate the end-to-end resource to be reserved, but the datalink layer actually has responsibility for reservation and QoS control. Current datalink media that can control QoS are ATM, PPP over SDH/SONET with packet scheduler, serial line with packet scheduler, IEEE 802.9 isoEthernet, IEEE 802.12 100VG-Any LAN, and Ethernet/Token Ring with IEEE 802.1p. The I/O channel media are IEEE 1394 and Universal Serial Bus. Inside the user site, these media may be used. However, every node in the commercial resource reservation service backbone will have to handle between 1,000 and 10,000 IP flows initiated by users. To handle such a large volume of IP flow, the IP flow must be processed by hardware on a high-speed line interface such as SDH/SONET. The SDH/SONET hierarchy is defined as being a four times rate, 155M, 622M, 2.4G, and 10G bps. However, in an actual network, there is a strong possibility that a 622 Mbps line cannot handle all of the traffic, but there is no demand for a 2.4 Gbps line speed. In the commercial WAN environment, the intermediate-speed link, which is not supported by SDH/SONET, is provided by the ATM connection on the SDH/SONET. So, the use of ATM enables economical networks that can meet the traffic demand. After all, if, and only if, PPP over SDH/SONET is implemented by hardware, there is a possibility that speeds equivalent to ATM speeds can be attained over SDH/SONET, but economically this is entirely inappropriate for commercial network service. If SVC ATM is used as the datalink media of the resource reservation service, it may be possible to solve the problem that prevents billing based on distance without needing a database, that can extract information about distance for billing from the destination IP address. In this case, the E.164/NSAP, which are layered structure addresses and which contain information about provider and region, can be used. Therefore, ATM is an indispensable implementation technology for the commercial resource reservation service, and there is no room for any other selection. The remainder of this section examines an implementation method of the resource reservation setup protocol based on ATM. Suzuki Expires May, 1997 [Page 17] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 4.2 Mapping of ATM VC and IP Flow An IP flow and an ATM VC must be mapped one-to-one, and must not aggregate multiple IP flows to single ATM VC, because ATM itself has the traffic control function. It is only wasteful to control both the ATM-layer and the IP-layer traffic, and has no effect other than to raise equipment cost and increase processing overhead. An IP flow also must not be divided among multiple ATM VCs. Synchronization between multiple ATM VCs to the same destination is not guaranteed. Therefore, synchronized information would be needed between ATM VCs, and this causes unnecessary overhead, decreases the reliability of the IP flow, and wastes network resources. 4.3 Dynamic QoS Changes The need to change QoS in an ATM system is a serious problem. Part of the ITU-T SCS2 signaling protocol, Q.2963.1 specifies a procedure to change the peak rate, but this is not sufficient as a procedure to change QoS. Also, signaling procedures in ATM Forum UNI 3.1/4.0 do not support QoS changes. ITU-T traffic control recommendation I.371 and ATM Forum UNI 3.1/4.0 do not specify traffic behavior during QoS changing. The only possibility is ABT, which is defined in I.371 and supports QoS changes initiated by F5 OAM. Unfortunately, the current ABT supports point-to-point connection only. Therefore, current ATM standards cannot support QoS changes in practice. There is a possibility that QoS changes can be supported in the resource reservation service by releasing old ATM connections and establishing new ones, but this method causes the following problems. o Network overload may be caused by call processing. o Since call processing is closely related to ATM billing, the billing system complexity is increased. o New calls are not always successfully established. It is also feasible to establish new ATM connections and then release old ones, but this method causes the following problems. o The current ATM NIC for WS/PC does not support a large number of QoS VCs. o Although it is only temporary, old and new VCs exist simultaneously, which increases the complexity of the billing Suzuki Expires May, 1997 [Page 18] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 system. o New calls are not always successfully established. Especially, there is a possibility that only part of point-to-multipoint call will be reestablished successfully. This increases the complexity of exception processing. Therefore, QoS changes by signaling are impractical under current ATM standards, but should be supported after ATM standards specify it. ITU-T SG11 and SG13 and ATM Forum SIG SWG and TM SWG should specify a procedure for dynamic QoS changes and traffic behavior during QoS changing. Additional research on point-to-multipoint ABT is also needed. 4.4 Heterogeneous QoS Heterogeneous QoS, which not always the same at each receivers, is not supported by ATM point-to-multipoint connection. Originally, the concept of heterogeneous QoS was introduced to support hierarchical coded IP flow. But as described in [6], it should be supported by a presentation layer, and it is not responsible for the network and transport layers. In single-node operation, IP packet discarding based on contents is needed to support an IP flow that has heterogeneous QoS. But the network and transport layer cannot support such processing. When multiple receivers require different QoS, it does not make sense to establish a point-to-multipoint VC based on the largest QoS request. However a point-to-multipoint VC is established based on the maximum QoS, when a receiver that requests a larger QoS joins the multicast tree, or when the receiver that requested the largest QoS leaves the tree, and the QoS of the VC must be changed. However, current ATM standards cannot support such QoS changes. Also it does not have any advantage in terms of the tariff. The billing for a resource reservation service may be based on the QoS acquired in the ATM layer, which may be larger than the QoS requested in the IP layer. Therefore, to use the QoS from the ATM layer is obviously better for the IP flow. As explained above, current ATM standards cannot support heterogeneous QoS, and it does not offer any advantage in terms of tariff. Suzuki Expires May, 1997 [Page 19] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 4.5 Filter Spec ATM does not support a function, that merges information from various VCs, such as filter spec. It should also be supported by a presentation layer, and is not responsible for the network and transport layers. 5. Conclusions This document clarified the followings in terms of an architecture for the resource reservation setup protocol. o The basis of the resource reservation setup protocol is sender initiation. That is, the sender explicitly designates the receiver address, sends a resource reservation setup message and constructs a multicast tree. o In the case of receiver initiation, the receiver explicitly sends a joining request message; if the sender accepts it, the sender sends a resource reservation setup message to the receiver. o However, if the application is open to the public, an intermediate node inside the multicast tree decides whether to accept or refuse the joining request, and may send a resource reservation setup message as a response to the joining request message. ATM is an indispensable implementation technology for the commercial resource reservation service, and the following were clarified in terms of an implementation method of the resource reservation setup protocol based on ATM. o An IP flow and an ATM VC must be mapped one-to-one. o In practice, current ATM standards cannot support dynamic QoS changes. Such changes should be supported after the ATM standard specifies it. o Current ATM standards cannot support heterogeneous QoS, which also does not have any advantage in terms of tariff. The hierarchical coded IP flow should be supported by the presentation layer. o Filter spec should be supported by the presentation layer. Suzuki Expires May, 1997 [Page 20] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 Finally, if the billing policies of ISPs are fundamentally different from each other in the commercial resource reservation service, it will be difficult to achieve smooth interconnection. Therefore, the author believes that ISPs need to conclude agreements or to clarify recommendations concerning minimum common billing policies for the resource reservation service, especially on the definition of distance for billing purpose. 6. Security Considerations Security considerations are not discussed in this document. References [1] S. Herzog, "Accounting and Access Control Policies for Resource Reservation Protocols," Internet Draft, June 1996, . [2] R. Braden Ed., "Resource ReSerVation Protocol (RSVP)-Version 1 Functional Specification," Internet Draft, October 1996, . [3] L. Delgrossi and L. Berger, Ed., "Internet Stream Protocol Version 2 (ST2) Protocol Specification - Version ST2+," RFC 1819, August 1995. [4] S. Berson and L. Berger, "IP Integrated Services with RSVP over ATM," Internet Draft, September 1996, . [5] M. Borden and M. Garrett, "Interoperation of Controlled-Load and Guaranteed-Service with ATM," Internet Draft, September 1996, . [6] K. Sebayashi and H. Uose, "ATM Multicast Communications Method with Receiver-initiated QoS Guarantee," Global Information Infrastructure Evolution, ISBN 90-5199-290-4, IOS Press, 1996. Suzuki Expires May, 1997 [Page 21] INTERNET DRAFT draft-suzuki-res-resv-svc-arch-00.txt November, 1996 Acknowledgments This document is based on my discussions with many colleagues at NTT. I would like to specially thank Hiroshi Ishikawa, Sadahiko Kanou, Masaru Nishi, Satoshi Takamatsu, and Hideaki Arai of the NTT Network Strategy Planning Dept., and also Hisao Uose of the NTT Multimedia Networks Labs. for their valuable comments. Also section 4 of this document is based on various discussions during NTT Multimedia Joint Project with NACSIS. I would like to thank Prof. Shoichiro Asano of National Center for Science Information Systems for his invaluable advice in this area. Author's Address Muneyoshi Suzuki NTT Multimedia Networks Laboratories 3-9-11, Midori-cho Musashino-shi, Tokyo 180, Japan Phone: +81-422-59-2119 Fax: +81-422-59-3203 EMail: suzuki@nal.ecl.net Suzuki Expires May, 1997 [Page 22]