Postel, Internet Protocol, Network Working Group Request for Comments 791, September 1981.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....are needed and, if G is a hypercubic network such as a butterfly network, Omega Gamma, 2 n) bit headers are still needed. For these networks, the lower bound matches the upper bound achieved by a simple protocol that uses packet headers to keep track of the number of hops packets have traveled [P81]. This protocol applies to both the static failure model, as well as the dynamic failure model. For other network topologies, we develop new protocols that can be adapted to both fault models, and use substantially smaller headers. For example, if the network has a known feedback vertex set of ....

....there is a path of operational links from S to R in an n node network, then there is a simple path of operational links of length at most n Gamma 1. Thus, it suffices to use packet headers that count the number of intermediate processors the packet has passed through (including duplicate visits) [P81]. To send a message to R, processor S sends a packet with header 0 containing the message to each of its neighbors. When an intermediate node receives a packet from a neighbor and the packet header is less than n Gamma 2, the processor increments the header and forwards the packet to all its ....

Postel, Internet Protocol, Network Working Group Request for Comments 791, September 1981.

....to deliver an input sequence of messages at one site to another site but makes no guarantees about delivery of packets except not to mutate or duplicate. TCP is a standard example of a virtual circuit. TCP is built on top of IP which provides the functionality of a datagram service [Pos81a, Pos81b] A protocol for STP, such a TCP, is often called an end to end protocol because it is only implemented on the remote sites and relies on an underlying network protocol to provide the necessary services for full communication. The standard approach to solving STP is for the sending process to ....

....order from the sequence numbers provided that the protocol provides adequate error control to resend packets that may have been lost. Since packets, and hence their headers, must have bounded length, practical protocols use bounded sequence numbers. For example, TCP uses 32 bit sequence numbers [Pos81b] Traditional bounded sequence number protocols alone do not provide reliable communication in the presence of arbitrarily reordered packets. For this reason, the underlying network protocol IP uses the hop count mechanism to purge old packets from the network, thereby allowing TCP to use ....

J. Postel. Transmission control protocol. Network Working Group Request for Comments 793, September 1981.

....attempts to deliver an input sequence of messages at one site to another site but makes no guarantees about delivery of packets except not to mutate or duplicate. TCP is a standard example of a virtual circuit. TCP is built on top of IP which provides the functionality of a datagram service [Pos81a, Pos81b] A protocol for STP, such a TCP, is often called an end to end protocol because it is only implemented on the remote sites and relies on an underlying network protocol to provide the necessary services for full communication. The standard approach to solving STP is for the sending ....

....do not provide reliable communication in the presence of arbitrarily reordered packets. For this reason, the underlying network protocol IP uses the hop count mechanism to purge old packets from the network, thereby allowing TCP to use bounded sequence numbers and provide reliable communication [Pos81a] Without a method of guaranteeing the removal of old packets from the network, bounded sequence numbers could wrap around, thereby allowing a bounded sequence number protocol to fail. Our protocols to solve STP are based on theoretical protocols for STP developed by Attiya et al. AFWZ89] and ....

J. Postel. Internet protocol. Network Working Group Request for Comments 791, September 1981.

....message using headers with fewer than dlog 2 ne 3 bits, generates an in nite amount of message trac. If there is a path of live links from S to R in an n node network, then there is a simple path of live links of length at most n 1. Therefore, it suces to use the simple hop count algorithm [P81] which discards messages that have been forwarded n 1 times. Since this can be done with headers of size dlog ne, for the complete graph we have upper and lower bounds that match to within a small additive constant, and for the n 2 input butter y and the n n 2 mesh to within a small ....

J. Postel, Internet Protocol, Network Working Group Request for Comments 791, September 1981. 12

....message trac. If there is a path of live links from S to R in an n node network, then there is a simple such path of length at most n 1. Therefore, it suces to use the simple hop count algorithm: use packet headers of length dlog 2 (n 1)e to count the number of links that packets have travelled [P81]. Thus, for the complete graph, we have upper and lower bounds that match to within a small additive constant, and, for the n 2 input butter y and the n n 2 mesh, to within a small multiplicative constant. However, for several graphs there remains a large gap between the best upper and ....

Postel, Internet Protocol, Network Working Group Request for Comments 791, September 1981. 15

....3 are needed and, if G is a hypercubic network such as a butterfly network, Omega Gammak, n) bit headers are still needed. For these networks, the lower bound matches the upper bound achieved by a simple protocol that uses packet headers to keep track of the number of hops packets have traveled [P81]. This protocol applies to both the static failure model, as well as the dynamic failure model. For other network topologies, we develop new protocols that can be adapted to both fault models, and use substantially smaller headers. For example, if the network has a known feedback vertex set of ....

....there is a path of operational links from S to R in an n node network, then there is a simple path of operational links of length at most n Gamma 1. Thus, it suffices to use packet headers that count the number of intermediate processors the packet has passed through (including duplicate visits) [P81]. To send a message to R, processor S sends a packet with header 0 and the message to each of its neighbors. When an intermediate node receives a packet from a neighbor and the packet header is less than n Gamma 2, the processor increments the header and forwards the packet to all its other ....

Postel, Internet Protocol, Network Working Group Request for Comments 791, September 1981.

....an ARP binding though such mechanisms have been proposed [13] Routing daemons bind routes to interfaces. Routes are often cached in protocol control blocks to avoid searching the routing table for each packet. Connections are another type of binding association. Transport protocols like TCP [22] represent a connection as a binding between two endpoints. These endpoints are usually network addresses, which are in turn bound to interfaces. Other possible bindings include quality of service (QoS) agreements, scope, group management (IGMP) and security. While these assumptions are not ....

Postel, J. Transmission Control Protocol. Network Working Group Request for Comments: 793, September 1981.

.... of packets (in networks with cycles) each intermediate processor also remembers the largest sequence number it has seen and only forwards packets with that sequence number [8, 27] An alternative method to limit the number of packets in the network is to add a hop count to the packet headers [22]. Each time a packet is forwarded along an edge, its hop count is incremented. A packet is destroyed whenever its hop count exceeds the number of processors in the network or the length of the longest simple path between S and R. Instead of counting the number of hops a packet has travelled, it ....

J. Postel, Internet Protocol, Network Working Group Request for Comments 791, September 1981.

....IBM, DEC, or Intel. Keywords: Search, Browsing, Information Retrieval, Distributed Systems, File Systems, Prefetching, Latency, Dynamic Sets 1. Introduction The last few years have seen the emergence of a new class of data repository, world wide information systems (WWIS) WWIS, such as FTP[19], WWW[1] Gopher[15] and world wide AFS[25] consist of loose collections of data repositories spanning the globe, and store massive amounts of data. The vast amount of accessible data makes search a critical application for these systems. Unfortunately, accessing data on remote machines has an ....

J. Postel and J. Reynolds. File transfer protocol (FTP). Network Working Group Request for Comments (RFC) 959, ISI, October 1985. Available as http://info.cern.ch/hypertext/WWW/Protocols/rfc959/Overview.html.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC