The TCP/IP is defined IETF 1981. The constitute one of the most widely used protocol suits I the computer communication area. A very large number of workstations, hosts, and even PCs are interconnected using TCP/IP. The Internet and especially the WWW, which is also based on IP, are very popular today. Many subscribers throughout the world ‘surf’ the Internet, which has exponential growth rates. Hence ATM-based networks must also be capable of carrying and supporting these protocol. In ATM networks TCP/IP protocol will be placed on top of ATM.

If native ATM applications using the UBR or ABR transfer capabilities are developed they might build the Internet of the future. However, this will definitely take some more time and thus interworking with TCP/IP-based networks will be essential for the next decade.

IP is a connectionless protocol designed for data applications. It can be run over various link-layer transmission systems like Ethernet, token ring and leased lines. The IP addresses are subdivided into a network IP and a host ID. IP datagrams contain the source and destination addresses. Sometimes, for an IP address the corresponding HW address is not known. In this case the address resolution protocol (ARP) will be used: the sender broadcasts an ARP request message to all stations. A station recognized its own IP address within this ARP request will send back an ARP reply containing its HW address. This enables the source to send data to the designated destination.

IP packets are transferred in a vest-effort manner. This means that no error correction capabilities, no QoS guarantees and no unknown destination address the internet control message protocol (ICMP) can be used inform the sender. ICMP messages are encapsulated and conveyed in IP packets.

IP data packets are transported in an unassured manner without any guarantee that they will arrive in the same order as sent out. Therefore, high-layer protocols, such as TCP, have to ensure data integrity. TCP is a connection-oriented reliable transport protocol, whereas the user datagram protocol (UDP) is a connectionless unreliable transport protocol. On top of these protocols, applications like simple network management protocol (SNMP), file transfer protocol (FTP), simple mail transfer protocol (SMTP), and remote login (TELNET) can be found.

When discussing IP over ATM three main aspects need to be considered:

encapsulation, routing and addressing/address resolution. In the following the basic principles of carrying IP traffic over an ATM-based network are described. These mechanisms are based on the following Internet Engineering Task Force specifications, called requests for comment (RFCs).

The most mature mechanism is based on RFC 1577, its draft update and other associated RFCs, this mechanism is also known as classical IP and ARP over ATM and further elaborated in the following.

This model considers ATM to be a ‘wire’ for the transport of higher-layer protocol information. The concept is similar to LANE, but uses a different addressing level, IP, in stead of the medium access control (MAC) level. It describes the transmission of IP datagrams, their encapsulation into ATM cells and the ATM address resolution protocol (ATM-ARP).

In order to reflect the fact that the IP network on top of ATM builds a logical association of network associations of network elements, such as hosts or routers, which is independent of the underlying ATM network, the notion logical IP subnetwork (LIS) was introduced; ‘an LIS consists of a group of IP nodes that connect to a singe ATM network and belong to the same IP subnetwork’.

The model is called ‘classical’ because the classical concept of subnetwork is retained. Each subnetwork can be of any type, for example frame relay, ethernet of ATM. Within a subnetwork, namely an LIS, direct connections between stations are possible and connections towards a station outside an LIS must be conveyed via a default router. This router would consistute an ATM endpoint and can belong to several LISs via a single physical interface.

However, such a router could easily become a bottleneck and end-to-end connections with a guaranteed QaS cannot be established.

Within an LIS all ATM connections used for information transfers are VCCs which can be either permanent of switched connections.

The maximum transmission unit (MTU) size within an LIS is the same for all VCs. The default value has been specified as being 9180bytes as in the ‘IP over SMDS’ case. Together with the 8 octets LLC/SNAP header this results in an AAL 5 PDU size of 9188 octets.

The classical IP and ARP over ATM approach requires that connection between hosts in different LISs are established via router hops. Since every router constitutes a potential bottlenecks and prevents the use of a transparent ATM end-to-end connection. The next hop resolution protocol (NHRP) allows a cut-through connection through several LISs to be established. Only the establishment of a direct and transparent ATM connection end-to-end allows, for example, ATM’s particular QoS capabilities to be exploited.

The next hop server (NHS) performs the next hop resolution protocol and realized an inter-LIS ARP mechanism that can be used by a source host to establish such a ‘cut-through’ connection. If the destination host is not park of the ATM network the NHS provides the address of the egress router through which this destination can best be reached. Any host can use the NHRP or router attached to an ATM network. Coexistence with the existing, classical ARP mechanisms are possible.

Although NHRP is described here in the context of IP and ATM networks, generally it can be used for other networks as well. Therefore, it has been adopted by the ATM Forum as the basis for carrying multiple protocols over ATM, that is , MPOA. The use of NHRP between two routers is proposed in (IETF ,1996i). Furthermore, it is likely that this protocol will be adopted for use in classical IP networks as well.

The NHRP as such is not a routing protocol but a specific ARP mechanism that avoid extra IP hops. Its address resolution is based on routing tables created either manually or by a dynamic routing protocol. The manually created, static routing tables, of course, are only applicable to small network configurations. The dynamic protocols are assumed to create a loop-free routing table.

Up to now the ‘classical model has not tacked the problem of how to support the connectionless IP multicast service via a connection oriented ATM network. Before describing the multicast/broadcast support mechanisms the following terms should be introduced:

unicast traffic : an information flow destined for a single, specific receiver;

multicast traffic : an information flow sent to a specific group of receivers;

broadcast traffic : information sent to all destinations within a specific networking environment.

With IP over a shared medium like ethernet a source can easily sent information to a multicast group address. The source ‘broadcast’ its message via the shared medium to all attached stations. Each individual station then decides for itself if it wishes to join this multicast group. The different groups are identified message. Individual hosts perform joining or leaving a group, their decision has no impact on any other hosts attached to the same network.

The need for supporting IP in large-scale, high throughput and reliable transport networks has recently led to a variety of proposals on how best to deal with IP traffic. Common to all these ideas is to speed up layer-3 routing by layer-2 routing (‘layer’ refers to the OSI reference model).

Most of these proposals originate from company-specific solution. Some of them are largely based on existing standards, while attempts are being made to standardize others through the IETF’s multiprotocol label switching (MPLS) working group; a first draft of RFC ca be found in IETF, 1997a.

IP switching    : a proposal from Ipsilon, described in several information RFCs

Tag switching : a Cisco solution which is currently studied in the MPLS working group

Carrier scale internetoworking : the Siemens/Newbridge solution for IP traffic, largely based on                          existing standards.

reference - www.atmforum.com