![typically required implementation in software or ®rmware. Category 2 can be
further classi®ed into the overlay model and the peer model.
In the overlay model, ATM switches are not aware of IP addresses and IP
routing protocols. This model overlays an IP network onto an ATM network,
essentially creating two network infrastructures with two addressing schemes and
two routing protocols. Each end system uses both IP and ATM addresses that
are uncoupled. Thus an address resolution protocol is required to map from one
address to another. One advantage of this model is that the ATM infrastructure
can be developed independently of the IP infrastructure. Examples of this model
are classical IP over ATM and multiprotocol over ATM.
The peer model uses the existing IP addresses (or algorithmically derived
ATM addresses) to identify end systems and uses IP routing protocols to set
up ATM connections. One advantage of the peer model is that it does not require
an address resolution protocol to interwork routeable address spaces and thus
simpli®es address administration. A node typically has an integrated ATM
switching and IP routing function, and the node can be viewed as a ``peer'' to
other routers. The peer model maintains one network infrastructure. The best
example of this model is multiprotocol label switching.
10.2 OVERLAY MODEL
IP is the dominant internetworking layer, while ATM is perceived as an econom-
ical switching solution for high-speed backbone networks. For this reason, there
has been much interest in overlaying IP internetworking protocol on top of
ATM. In this section we look at three IP-over-ATM approaches.
10.2.1 Classical IP Over ATM
The classical IP over ATM (CLIP) model [RFC 2255] is an IETF speci®cation
whereby IP treats ATM as another subnetwork to which IP hosts and routers are
attached. In the CLIP model multiple IP subnetworks are typically overlaid on
top of an ATM network. The part of an ATM network that belongs to the same
IP subnetwork is called a logical IP subnetwork (LIS), as shown in Figure 10.2.
All members (IP end systems) in the same LIS must use the same IP address
pre®x (e.g., the same network number and subnet number). Two members in the
same LIS communicate directly via an ATM virtual channel connection (VCC).
Each LIS operates and communicates independently of other LISs on the
same ATM network. Communications to hosts outside the LIS must be provided
via an IP router that is connected to the LIS. Therefore, members that belong to
different LISs must communicate through router(s).
Suppose a host (host S) wants to use CLIP to send packets to another host
(host D). When host S sends the ®rst packet to host D in the same LIS, host S
10.2 Overlay Model 677
| | | Textbook Table of Contentse-Text Main Menu
v
v](https://image.slidesharecdn.com/ipforwardingarchitecturesandoverlaymodel-200323104258/85/IP-forwarding-architectures-and-Overlay-Model-2-320.jpg)
IP forwarding architectures and Overlay Model
- 1. 6. Differentiated services. We describe the differentiated services model that has been introduced by IETF as a scalable approach to provide levels of QoS support in the Internet. 10.1 IP FORWARDING ARCHITECTURES The Internet has been growing at an exponential rate in terms of the number of hosts and domains, and traf®c demand. This high growth in traf®c demand has been stressing existing infrastructures in the core networks. To maintain their network performance and avoid congestion collapse, service providers have been constantly upgrading their backbone links, typically with ATM technology. When transmission links are upgraded, routers and switches may have to be upgraded as well to keep up with increasing link speeds. This situation has led to changes in router architecture to extend the achievable performance. A num- ber of IP forwarding solutions have resulted, and they can be categorized as shown in Figure 10.1. Category 1 retains the same forwarding paradigm as the conventional router architecture but solves the potential bottlenecks in traditional routers by mod- ifying the internal architecture. For example, one change is the replacement of the bus backplane in the data paths with a switch backplane, thus providing simultaneous packet forwarding. In addition, scalability is improved by placing an IP lookup engine at each interface. This approach is incorporated in the design of gigabit routers. Category 2 simpli®es the lookup process by using short, ®xed-length labels rather than long, variable-length IP pre®xes. A typical way to simplify this process is to run IP over ATM, which uses the VCI and VPI in the lookup process. Because label lookup uses direct indexing, it can be easily performed in hardware. For example, label lookup in ATM simply uses the incoming VPI/ VCI value as a direct index to a connection table entry to determine the outgoing VPI/VCI, the output port, and other relevant information. ATM label lookup can be easily done in one cell time. On the other hand, IP address lookup has 676 CHAPTER 10 Advanced Network Architectures High-performance forwarder (2) Switched forwarding (a) Overlay model (b) Peer model (1) Pure destination-based forwarding (lookup based on IP address) FIGURE 10.1 IP forwarding taxonomy | | | Textbook Table of Contentse-Text Main Menu v v
- 2. typically required implementation in software or ®rmware. Category 2 can be further classi®ed into the overlay model and the peer model. In the overlay model, ATM switches are not aware of IP addresses and IP routing protocols. This model overlays an IP network onto an ATM network, essentially creating two network infrastructures with two addressing schemes and two routing protocols. Each end system uses both IP and ATM addresses that are uncoupled. Thus an address resolution protocol is required to map from one address to another. One advantage of this model is that the ATM infrastructure can be developed independently of the IP infrastructure. Examples of this model are classical IP over ATM and multiprotocol over ATM. The peer model uses the existing IP addresses (or algorithmically derived ATM addresses) to identify end systems and uses IP routing protocols to set up ATM connections. One advantage of the peer model is that it does not require an address resolution protocol to interwork routeable address spaces and thus simpli®es address administration. A node typically has an integrated ATM switching and IP routing function, and the node can be viewed as a ``peer'' to other routers. The peer model maintains one network infrastructure. The best example of this model is multiprotocol label switching. 10.2 OVERLAY MODEL IP is the dominant internetworking layer, while ATM is perceived as an econom- ical switching solution for high-speed backbone networks. For this reason, there has been much interest in overlaying IP internetworking protocol on top of ATM. In this section we look at three IP-over-ATM approaches. 10.2.1 Classical IP Over ATM The classical IP over ATM (CLIP) model [RFC 2255] is an IETF speci®cation whereby IP treats ATM as another subnetwork to which IP hosts and routers are attached. In the CLIP model multiple IP subnetworks are typically overlaid on top of an ATM network. The part of an ATM network that belongs to the same IP subnetwork is called a logical IP subnetwork (LIS), as shown in Figure 10.2. All members (IP end systems) in the same LIS must use the same IP address pre®x (e.g., the same network number and subnet number). Two members in the same LIS communicate directly via an ATM virtual channel connection (VCC). Each LIS operates and communicates independently of other LISs on the same ATM network. Communications to hosts outside the LIS must be provided via an IP router that is connected to the LIS. Therefore, members that belong to different LISs must communicate through router(s). Suppose a host (host S) wants to use CLIP to send packets to another host (host D). When host S sends the ®rst packet to host D in the same LIS, host S 10.2 Overlay Model 677 | | | Textbook Table of Contentse-Text Main Menu v v
- 3. knows only the IP address of host D. To set up a VCC, host S needs to know the ATM address of host D. How does host S resolve the ATM address of host D from the IP address? The solution is provided by implementing an ATM Address Resolution Protocol (ATM ARP) server on each LIS. The ATM address of the ATM ARP server is con®gured at each host. When a host boots up, it registers its IP and ATM addresses to the ATM ARP server on the same LIS. When a host wants to resolve the ATM address of another host from the IP address, the ®rst host asks the ATM ARP server for the corresponding ATM address. After the host receives an ATM ARP reply from the ATM ARP server, the host can establish a VCC to the destination host and send packets over the VCC. This process involves fragmenting the IP packet into ATM cells at the source host and the reassembly of the packet at the destination host. What happens if the destination host belongs to another LIS? In this situa- tion the source host simply establishes a VCC to the router connected to the same LIS. The router examines the IP packet, determines the next-hop router, establishes a VCC, and forwards the packet along to the next router. The process is continued until the router of the LIS of the destination is reached, and the packet is then delivered to the destination host. In CLIP, IP packets sent from the source host to the destination host in a different LIS must undergo routing through the LIS router, even if it is possible to establish a direct VCC between the two IP members over the ATM network. This requirement precludes the establishment of a VCC with a speci®c QoS between end systems. CLIP allows a permanent virtual connection (PVC) to be established between hosts of a LIS. In this case the connection is preestablished manually between two hosts. What one host needs to ®nd is the IP address of the other end. The host uses an inverse ATMARP (InATMARP) to ®nd the IP address of the other end. 678 CHAPTER 10 Advanced Network Architectures ATM network LIS1 LIS2 LIS3 LIS4 LIS5 LIS6 Router Router Router Router Router FIGURE 10.2 Classical IP over ATM model | | | Textbook Table of Contentse-Text Main Menu v v
- 4. 10.2.2 LANE LAN emulation (LANE) is an ATM Forum speci®cation intended to accelerate the deployment of ATM in the enterprise network. Typically, a host runs an internetwork layer protocol such as IP over a ``legacy'' LAN such as Ethernet or token ring. LANE enables any software that runs on a legacy LAN to also run on an ATM network without any modi®cation. LANE works by presenting the network layer with an interface that is identical to that of legacy LANs. Figure 10.3 illustrates the changes in the lower layers such that the interface from the device driver to the network layer (e.g., NDIS1 ) remains unchanged. LANE maintains the same interface between the network layer and the data link layer, so in effect an ATM network can be made to appear like an Ethernet or token ring LAN to the higher layer software. This behavior also enables LANE to support other network layer protocols such as IPX and AppleTalk. In con- trast, the CLIP model supports IP only. An emulated LAN (ELAN) consists of the following components (see Figure 10.4): A set of LAN emulation clients (LECs) LAN emulation server (LES) Broadcast and unknown server (BUS) LAN emulation con®guration server (LECS) A LEC resides in the end system (e.g., host, server, bridge, etc.) and performs data forwarding, address resolution, and control functions. Each LEC is identi- ®ed by a unique ATM address. A LES responds to LEC address resolution requests by resolving MAC addresses to ATM addresses. A BUS handles broad- cast, multicast and initial (i.e., before a VCC is established) traf®c in a given ELAN. One main purpose of the LAN Emulation Con®guration Server (LECS) is to assign LECs to the corresponding ELANs (i.e., associate a LEC to the correct LES). During the registration phase, each LEC noti®es the LES of its ATM and MAC addresses. When a LEC (say, LEC1) wants to send a frame to another 10.2 Overlay Model 679 1 NDIS stands for network driver interface speci®cation. Microsoft networking protocols interact with network card drivers by using NDIS. NDIS operates at the logical link control sublayer of the data link layer. NDIS allows the binding of multiple NDIS-compliant NIC cards with one protocol stack, multiple protocols with a single NIC card, or multiple protocols with multiple NICs. Network layer LLC LLC LANE AAL5 ATM MAC Network layer FIGURE 10.3 Legacy LAN and LANE protocol stacks | | | Textbook Table of Contentse-Text Main Menu v v
- 5. LEC (say, LEC2), LEC1 ®rst checks if it knows the ATM address of LEC2. If LEC1 does not know the ATM address, then it sends an LE_ARP request to the LES. In the meantime, LEC1 sends subsequent frames via the BUS. If there was an earlier registration from LEC2, the LES can resolve LEC2's MAC address to the ATM address. After receiving the LE_ARP reply, LEC1 will cache the ATM address of LEC2 and set up a VCC to LEC2. From then on frames from LEC1 to LEC2 are transmitted through the VCC. The cache is aged out so that inactive VCCs are eventually purged. LANE has several shortcomings. LANE, by virtue of operating at the MAC layer, is susceptible to broadcast storms. The requirement that LANE hide the details of the underlying ATM network from the network layer also implies that the QoS attributes of ATM cannot be made available to the network layer protocols. LANE is de®ned to operate over UBR and ABR connections, which more closely match the service provided by LAN MAC protocols. 10.2.3 NHRP LANE enables a station to resolve an ATM address from a MAC address. The Next-Hop Resolution Protocol (NHRP) enables a station connected to an ATM network to resolve an ATM address from an IP address. NHRP allows a host to determine the ATM address of another host or of an egress router from the ATM network. The main objective of NHRP is to ®nd the most ef®cient shortcut path through the ATM network so that intermediate routers can be bypassed. Recall that the CLIP model resolves only the ATM address that belongs to the same LIS. In other words, the CLIP model requires a router to perform packet forwarding between two different LISs. In contrast, NHRP allows a shortcut to traverse multiple LISs, making it more suitable for larger networks. Figure 10.5 illustrates the key difference between the path generated by classical IP (default path or routed path) and the path generated by NHRP (shortcut path or cut-through path). NHRP is based on a client/server architecture. An NHRP cloud contains entities called next-hop clients (NHCs), which are responsible for initiating NHRP resolution request packets, and next-hop servers (NHSs), which are 680 CHAPTER 10 Advanced Network Architectures ATM network LEC LEC LEC LEC LES BUS LECS FIGURE 10.4 LANE con®guration | | | Textbook Table of Contentse-Text Main Menu v v