📄 rfc1221.txt
字号:
Network Working Group W. EdmondRequest for Comments: 1221 BBNUpdates: RFC 907 April 1991 Host Access Protocol (HAP) Specification - Version 2Status of this Memo This memo describes the Host Access Protocol implemented in the Terrestrial Wideband Network (TWBNET). It obsoletes most but not all of RFC 907. This memo provides information for the Internet community. It does not specify an Internet standard. Distribution of this memo is unlimited.Preface This memo specifies the Host Access Protocol (HAP). HAP is a Network layer (OSI Layer 3 lower) access protocol that was first implemented about a decade ago for the DARPA/DCA sponsored Wideband Packet Satellite Network (WBNET), the precursor of the current Terrestrial Wideband Network (TWBNET). This version of the specification obsoletes references [1] and [2] in addition to most of RFC 907. HAP is a developmental protocol, and will be revised as new capabilities are added and unused features are eliminated or revised. One reason that HAP is being revised now is that, unlike the original WBNET's satellite channel, the TWBNET's T1 fiber links are not a broadcast medium. This has prompted some changes to the protocol that will permit greater efficiency in a mesh topology network. Another cause of revision is the need to make HAP able to support a variety of OSI layer 3 upper protocols, such as DECNET Phase V, ST, and CLNP, where before only Internet Protocol (IP) was used. Appendix B describes how backward compatibility with the older IP- only version of HAP is achieved. A third cause of protocol changes is the desire to simplify interaction between ST2 protocol (RFC 1190) agents and the TWBNET. This has mainly affected the way certain setup errors are handled. These changes are expected to be backward compatible. Appendix A describes two capabilities that may be added to HAP in the future. One of the protocol enhancements, "Group Streams", described in reference [2] has been eliminated. There are no known applications that use the feature. As described in Appendix A, a new mechanism, to be called "shared streams", capable of providing equivalent capabilities will be implemented if needed. Changes in [2] that have been retained include various query/reply control messages that permit a host to determine what resources it owns (mostly useful forEdmond [Page 1]RFC 1221 HAP2 April 1991 cleanup following a host reboot or crash). This document assumes the reader is familiar with DoD internetworking terminology.1. Introduction The Host Access Protocol (HAP) is a network layer protocol (as is X.25). ("Network layer" here means ISO layer 3 lower, the protocol layer below the DoD Internet Protocol (IP) layer [3] and above any link layer protocol.) HAP defines the different types of host-to- network control messages and host-to-host data messages that may be exchanged over the access link connecting a host and the network packet switch node. The protocol establishes formats for these messages, and describes procedures for determining when each type of message should be transmitted and what it means when one is received. HAP has been implemented in the wide-area network called the Terrestrial Wideband Network (TWBNET) [5] and in the routers and other hosts that connect to TWBNET. The packet switch nodes that compose the TWBNET are called Wideband Packet Switches (WPS). Both the precursor to HAP, the Host/SATNET Protocol [6], used in the Atlantic Packet Satellite Network (SATNET) and the Mobile Access Terminal Network (MATNET [7]), and HAP, used in the original Wideband Satellite Network (WBNET) [8], were originally designed to provide efficient access to the single satellite channel each network used to connect all sites. The HAP protocol designers reflected some of the peculiarities of the single satellite channel environment in the HAP protocol itself. The current Terrestrial Wideband Network (TWBNET) utilizes T1-speed fiber connections between sites. Future networks and TWBNET may use a combination of terrestrial connections and satellite connections, and may have more than one of each. The HAP protocol has been changed to accommodate these extensions. Section 2 presents an overview of HAP. Details of HAP formats and message exchange procedures are contained in Sections 3 through 10. Further explanation of some of the topics addressed in this HAP specification can be found in reference [1]. Any protocol employed to provide sufficiently reliable message exchange over the Host-WPS link is assumed to be transparent to the protocol defined in this document. Examples of such link-level protocols are ARPANET 1822 local and distant host [9], ARPANET VDH protocol [9], and HDLC.Edmond [Page 2]RFC 1221 HAP2 April 19912. Overview HAP can be characterized as a full duplex, nonreliable protocol with an optional flow control mechanism. HAP messages flow simultaneously in both directions between the WPS and the host. Transmission is nonreliable in the sense that the protocol does not provide any guarantee of error-free sequenced delivery. If error-free delivery on the host's access link is required, it must be provided by the link layer protocol below HAP. (Use of link layer protocols for this purpose is not within the scope of this document.) HAP's flow control mechanism operates independently in each direction, but the choice to enable flow control or not applies to both directions together. HAP supports host-to-host communication in two modes corresponding to the two types of HAP data messages, datagram messages and stream messages. Each type of message can be up to 2048 octets in length. The basic transmission service in the network is datagram service. Datagrams are variable length, unsequenced, independent, and delivery is not guaranteed. The HAP header of each datagram determines the processing of the message. On this datagram service base a "stream" service is built. Stream service provides network bandwidth guarantees, but requires explicit setup and teardown operations to allocate and deallocate network resources. Stream traffic is best suited for continuous media traffic, but may also be used to obtain the lowest possible network delay. Host streams are established by a setup message exchange between the host and the network prior to the commencement of data flow. Although established host streams can have their characteristics modified by subsequent setup messages while they are in use, the fixed allocation properties of streams relative to datagrams impose rather strict requirements on the source of the traffic using the stream. Stream traffic arrivals must match the stream allocation both in interarrival time and message size if reasonable efficiency is to be achieved. The characteristics and use of datagrams and streams are described in detail in Sections 3 and 4 of this document. Both datagram and stream transmission in the network use logical addressing. Each host on the network is assigned a permanent 16-bit logical address which is independent of the physical port on the WPS to which it is attached. These 16-bit logical addresses are present in all Host-to-WPS and WPS-to-Host data messages. HAP supports multicast addressing via "groups". Multicast addressing is provided primarily to support the multi-destination delivery required for conferencing applications. Group addresses areEdmond [Page 3]RFC 1221 HAP2 April 1991 dynamically created and deleted by the use of setup messages exchanged between a host and the WPS. Membership in a group may be any arbitrary subset of the network hosts. A message addressed to a group address is delivered to all hosts that are members of that group, except the sender. Once a multicast address has been created, any member host may use that address, not just the creator. Although HAP does not guarantee error-free delivery, error control is an important aspect of the protocol design. HAP error control is concerned with both local transfers between a host and its local WPS and transfers through the network to the destination(s). The WPS offers users a choice of network error protection options based on the network's ability to selectively send messages over its transmission media at different forward error correction (FEC) rates. These FEC options are referred to as reliability levels. Four reliability levels (low, medium-low, medium-high, and high) are available. The precise error rate provided by each reliability level is not specified. Various checksum and CRC mechanisms are employed in the network to provide an error detection capability. A host has an opportunity when sending a message to indicate whether the message should be delivered to its destination or discarded if a data error is detected by the network. Each message received by a host from the network will have a flag indicating whether or not an error was detected in that particular message. A host can decide on a per-message basis whether or not it wants to accept or discard transmissions containing data errors. For connection of a host and WPS in close proximity, error rates due to external noise or hardware failures on the access circuit may reasonably be expected to be much smaller than the best network trunk circuit error rates. Thus for this case, little is gained by using error detection and retransmission on the access circuit. A 16-bit header checksum is provided, however, to ensure that WPSen do not act on incorrect control information. For relatively long distances or noisy connections, retransmissions over the access circuit may be required to optimize performance for both low and high reliability traffic. It is expected that link layer error control procedures (such as HDLC with retransmission) will be used for this purpose, but use of a reliable link layer protocol is not within the scope of this document. Each datagram message submitted to the WPS by a host is marked as being in one of three priority classes, from priority 2 (highest) through priority 0 (lowest). The priority class is used by the WPS for arbitrating contention for scarce network resources (e.g., link bandwidth). That is, if the network cannot deliver all of theEdmond [Page 4]RFC 1221 HAP2 April 1991 offered messages, high priority messages will be delivered in preference to low priority messages. Priority level affects the order of access to intersite link bandwidth and the order of message delivery at the destination WPS. Each stream message also has three priority classes, from priority 2 (highest) through priority 0 (lowest). In addition, streams themselves have three precedence classes, from precedence 2 (highest) through precedence 0. A stream of higher precedence can preempt a stream of lower precedence at setup time. Stream message priority provides a mechanism for a low-bandwidth host to receive a high- bandwidth stream and selectively discard messages marked as less important by the sender. Stream message priority does not affect the order of delivery of stream messages between the source and the destination. Datagram and stream messages being presented to the WPS by a host may not be accepted for a number of reasons: priority too low, destination dead, lack of buffers in the source WPS, etc. The host faces a similar situation with respect to handling messages from the WPS. To permit the receiver of a message to inform the sender of the local disposition of its message, an acceptance/refusal (A/R) mechanism is implemented. The mechanism is the external manifestation of the WPS's (or host's) internal flow and congestion control algorithm. If A/Rs are enabled, an explicit or implicit acceptance or refusal for each message is returned to the host by the WPS (and conversely). This allows the host (or WPS) to retry refused messages at its discretion and can provide information useful for optimizing the sending of subsequent messages when the reason for refusals is also provided. The A/R mechanism can be disabled to provide a "pure discard" interface. The host's choice to use the A/R mechanism or not does not limit its ability to send and receive messages to any other hosts. While the A/R mechanism allows control of individual message transfers, it does not facilitate regulation of priority flows. Such regulation is handled by passing advisory status information (GOPRI) across the Host-WPS interface indicating which priorities are currently being accepted. As long as this information, relative to the change in priority status, is passed frequently, the sender can avoid originating messages which are sure to be refused. HAP defines both data messages (datagram messages and stream messages) and link control messages. Data messages are used to send information between hosts on the network. Link control messages are exchanged between a host and the WPS to manage the local access link. Allocation of network resources, such as streams and groups, isEdmond [Page 5]RFC 1221 HAP2 April 1991 accomplished via an exchange of datagram messages, called Setups, between the user host and an agent inside the WPS called the "Service Agent." Setups are used to reserve, allocate, modify, free, and deallocate network resources. Each allocated resource has a unique identifier which, when placed in an appropriate field in a message header, allows that message to use the resource. E.g., after an exchange of Setups to create a group address, a message may be sent to the group by placing the group address in the destination field of that message. The Service Agent also permits a host to inquire about resources it owns.
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -