In the past, internetworking was realized by using bridges and routers. In fact the lack of intelligence in bridges and their weakness for controlling the broadcast traffic resulted in creation of routers. With the growth of networks, multiprotocol routers were born to interconnect LAN segments. They can control the access permissions, the broadcast traffic, and the traffic flow between segments.
Routers added more security and stability to networks but now they are too slow to handle the heavy load of router-based backbones. Bandwidth bottleneck, the high and increasing cost of the network operation, and increasing complexity of the network management of router-based networks are problems hard to overcome.
The traditional approaches, for solving the problems with growth were all based on shared media architecture. There is no question that networking technology should migrate from shared media to sw itching technology. The question is not wether the switching technology should replace the shared media technology or not. The issue is how the migration should be realized and how the existing installed base networks should transform to the new technology.
Improvements in hubs, bridges, and routers have increased the complexity of the internetworking in a way that adding, changing and moving of the endstations in the network results in unacceptable operation costs for internetworking in the traditional mode. Network growth and new applications require a change in paradigm.
Switch-based internetworking provides high performance and less complex networks that have lower operation costs. The switch based solution includes two approaches which are LAN switching and ATM switching. The LAN switching introduces the switching of the existing packet-based LAN technologies (such as Ethernet, Token Ring, and FDDI). The ATM is a cell-based switching technology based on connection oriented approach.
Considering the bandwidth demand of today's applications, the router based backbone can not be an appropriate solution any more. In a traditional router based backbone, the backbone traffic may prevent the multimedia applications from running or applications may become slow or even suffer from time-out.
Another problem with traditional router based backbones is their administration costs. Growth in networks and changes are a major part of the operation costs of a router-based network. The time consuming part of the administration of changes is the assignment of subnetwork addresses to the physical LAN segments. Subnetwork here means any logical subset of a network. It can be an IP subnet, an IPX network, DECnet, etc. For example in the case of an IP network, when an endstation moves from one LAN segment to another, the subnet address of its IP address should be changed. This results in a high operation cost.
FIGURE 1. Router as a collapsed backbone
FIGURE 2. Switch-based backbone
Internetworking is supposed to provide connectivity over LAN and WAN. In the case of traditional LANs, the forwarding is based on MAC layer or Network layer. In the case of Virtual LANs, the boundaries of virtual LANs define the routing requirements. In the case of ATM networks, the routing decision is made for the first cell in a flow and a connection is established. Then the rest of the cells in the flow are sent via simple switches that forward the cells simply based on their Virtual Connection Identifier (VCI) and Virtual Path Identifier (VPI).
Switches bring speed to the backbone and virtual networking removes the bottleneck of physical LANs. Asynchronous Transfer Mode (ATM) is a scalable solution that offers high speed, high performance, high availability and simplified administration. ATM provides dedicated bandwidth, different classes of service that can support the multimedia applications, bandwidth or time sensitive applications, and high traffic client/server configurations. This technology can solve the problem of congested router-based backbones.
The endstations on different LAN segments are connected via high speed switches without intermediate hops. Virtual networks can be realized in different ways. Virtual LANs are a layer 2 switched virtual network. The ATM LAN emulation can be realized by bridging the LAN segments. It can also realize the multiprotocol routing, although the corresponding standards are still to be defined and finalized. However, many implementations that realize routing based on the proprietary solutions are already in place.
When the number of LAN segments increases, ATM can be used as a backbone. High performance servers can be connected to the ATM backbone and WAN connection can be also realized via ATM. This is a complete solution that provides the network with high capacity, low and predictable delays, and easy administration.
Routers are slower than switches in nature because they should compute the route for each individual packet. This result in congestion and latency. However switches only check a small and simple header of the cell and pass the cell to the output. LAN switching boxes offer full bandwidth to all their ports. The end-stations are bridged at the MAC layer while an endstation can be connected directly to a port when it demands more bandwidth. In this case only one MAC is mapped to that specific port. Since this is a layer 2 implementation, it cannot offer a firewall or broadcast.
A logical group of endstations on different LAN segments build a virtual LAN which is basically a layer 2 connection. The virtual networks are connected to each other by using routers that provide the layer 3 networking.
FIGURE 3. LAN Switching (router backbone)
FIGURE 4. LAN Switching (ATM backbone)
When a LAN segment is congested and the endstations on that segment ask for more bandwidth, the number of endstations on the segment should be reduced. Thus the segment should be split into smaller segments or microsegments with less endstations per segment. The number of LAN segments causes no bottleneck for Virtual LANs.
In the case of router based networks, each physical LAN segment has a subnet address. When the LAN segmentation changes, the subnet addressing should change. As the number of segments increases, the network administration becomes more complicated. But in the case of Virtual LAN, physical segments are grouped in a logical subnetwork that has a subnet identification. This way physical segments can be grouped based on the departments, projects, etc. regardless of the physical configuration of the segments. This makes the administration simple. Endstations can move to different physical segments. There is no need to change the address (e.g. IP) of the endstations, while they are in the same Virtual LAN.
The endstations of a virtual LAN are bridged together while they can be on different LAN segments (hubs). The virtual LANs are connected to each other by routers. As a result, the routers can again become a bottleneck in the network when the traffic between virtual LANs increases. Moreover, the wiring between hubs and routers becomes a problem as the network grows. This leads to new solutions based on ATM technology.
However, when an endstation moves from one virtual LAN to the other, the network administer should manually change the network configuration to include the endstation in a different virtual LAN.
Using ATM as the backbone for LAN interconnection solves the congestion and reconfiguration problems mentioned above. ATM Forum's LAN Emulation (LANE) specification defines the required standard. It makes it possible to have an ATM backbone and attach LANs and ATM endstations to that backbone. As the LANE name stands, the traditional shared media LANs are emulated on the ATM backbone network and the ATM network is hidden from the view of the endstations on the LANs.
FIGURE 5. LAN Emulation Model
The translation of MAC address to ATM address is done by LAN Emulation Server. On the border of LAN and ATM networks, LANE Clients are located. Each LAN is connected to the ATM network through a LAN Emulation Client, which is responsible for address resolution (MAC to ATM address translation) and call setup based on User Network Interface (UNI) Q.2931. The broadcast server delivers the broadcast frames via the ATM backbone.
The assignment of physical segments to virtual LANs should be done manually. The routers used in such a network have an ATM UNI link and several LAN connection links.
ATM Edge routers integrate routing, ATM connection, and LAN switching in one device. They do the routing switching, and realizing the virtual subnets. The design goal of these routers is low latency to overcome the deficiency of traditional multiprotocol routers. Routing over ATM is addressed in IETF RFCs 1577 and 1483. Routing protocols like Open Shortest Path First (OSPF), Routing Information Protocol (RIP), and Interior Gateway Routing Protocol (IGRP) can be supported. The routing tables are updated using the standard ATM multicast services.
FIGURE 6. Edge Router Model
When an edge router wants to resolve an IP address, it uses the IP multicast channel to ask all the other edge routers about that address to find out the corresponding ATM address of the destination. The router establishes a virtual circuit to that edge router using the Q.2931 signalling. This basically routes the cells in ATM part of the network and the remote router sends the traffic to a router or the destination subnet and endstation. The LAN segments can be grouped into layer 3 virtual subnets that are based on different routable protocols. Such virtual subnets can have broadcast support and firewall implementation. However moving the endstations from one virtual subnet to another requires manual reconfiguration of the address.
The traffic management of virtual subnetsis easier than managing the physical ports or LAN segments. The administration applied to the virtual subnet layer is subsequently applied to the physical ports and segments via the intelligent software modules of edge routers.
The ARP and broadcast services might be provided by the same route server. Based on the network size and server load, there might be separate servers for routing, ARP and broadcast.
The switches, called multilayer switches, are the link between ATM backbone and LAN segments. Thus a virtual router consists of a route server and multilayer switches.
FIGURE 7. Virtual Router Model
This implementation is hidden from the endstations and the virtual router looks and behaves like a traditional physical router. The layer 3 packets have only one subnet hop and the ATM switches involved in this hop don't cause latency and congestion which is typical of the traditional router based networks.
In this concept, the traffic goes through the virtual circuit, which is set up in ATM backbone. The access restrictions, bandwidth allocation, and accounting can be all done in ATM network. The LAN-connected basestations can use all the ATM traffic management features.
The network administration becomes easier too. The administration is basically focused on the router server. The multilayer switches are not that complex and need simple administration through SNMP agents. In fact, since the multistage switches are much simpler than the multiprotocol routers, their cost per port is much lower. Network growth is simple too and the virtual routers are very scalable. However the port configuration should be done manually. When endusers move from one virtual subnet to another, the changes should be applied manually.
The standardization of protocols between multistage switches and route servers is still a barrier to this approach. However, the virtual router is supposed to be transparent to the network. From the point of view of other network elements, it should look like a physical router. All the existing traditional routers have their proprietary standards and protocols inside. As long as one thinks of a virtual router as a replacement for a traditional router, the virtual router can have its proprietary architecture inside. Hopefully the Private Network to Network Interface (PNNI) standard will be more extensive and will cover the virtual router protocols and signalling.
The relational LANs are the construction elements of a relational network. A relational LAN is a virtual LAN that supports a specific protocol and is automatically created in the network. In fact a relational LAN is a collection of endstations with the same subnetwork address and the same protocol. The endstations are automatically assigned to virtual LANs and a multiprotocol endstation will be the member of several virtual LANs.
A group of related users get a subnetwork address regardless of their physical point of connection to the network. This means a subnetwork is a logical identification for a user group, while in traditional networks, a subnetwork is a physical concept and usually a LAN segment. When an endstation moves from one LAN segment to the other and still in the same usergroup or relational LAN, it's still in the same subnetwork.
This definition can cover all routable protocols like DECnet, AppleTalk, and Netware and also nonroutable protocols like NetBIOS.
FIGURE 8. Relational Network (Physical)
FIGURE 9. Relational Network (Logical)
The relational LANs are connected to each other by using routers. As it was the case in the previous solutions, the endstations on the same relational LAN have zero hop distance regardless of their physical location on the network. Relational LANs, that represent user groups, are connected by routers.
The relational switch is the physical element of a relational network. Relational switches, which are switches with support for both LAN frames and also ATM cells, are connected by ATM links. The relational switch has the intelligence to interconnect related endstations and form a logical relational network. Such intelligence in software handles the administration tasks needed for reconfiguring the network. An endstation can be added or moved to any physical location and the relational switch can connect them based on their protocol type and logical subnetwork address.
The relational LAN is the logical element of a relational network. Those endstations that are grouped in the same relational LAN, have zero hop connection and the communication speed between them is as if they were physically connected to the same segment.
The relational LANs are formed based on the traffic of the network. For each existing subnetwork, a relational LAN is formed. For example in an IP network, each IP subnet has its relational LAN. Each non-routable protocol has a relational LAN. In the case that non-routable protocols exist on the same network, a relational LAN is formed for each non-routable protocol. If an endstation supports two different protocols, it will be in two different relational LANs. The subnetworks do not relate to a specific physical segment.
The physical configuration of the network is not a limitation for relational LANs. A relational LAN can be formed on multiple LAN segments and a LAN segment might be included in several relational LANs.
The routers are used to connect the relational LANs. Since the workgroups are mapped on the relational LANs, the traffic over the routers in minimal. Because routers connect the logical working groups regardless of the physical topology, it's easier to apply security and traffic control to the network.
Broadcast traffic is one of the important factors that affect the performance of the backbone. For routable protocols, the broadcast frames are sent only to those segments that have endstations with the same subnet address as the originator of the broadcast. In the case of non-routable protocols, the broadcast frames are sent only to those segments that have endstations with the same protocol as the originator's protocol. This results in better performance compared to the traditional bridging of non-routable protocols, which dictates unnecessary traffic to the backbone. The broadcast traffic can be further optimized too. For example, a Netware Service Advertisement Protocol (SAP) broadcast can be sent only to those segments that have Netware servers and routers. The segments with Netware clients or print servers are not flooded with such broadcasts.
Regarding the ATM physical layer, some of the available specifications are multimode and singlemode fiber 155 Mbps SONET STS-3C, multimode fiber and shielded twisted pair copper 155 Mbps 8B/10B coding, DS1 (1.5 Mbps), and DS3 (45 Mbps). Specifications for SONET/SDH 25.9, E1 (2 Mbps), and E3 (34 Mbps) and unshie lded twisted pair copper CAT-3 and CAT-5 are going to be completed.
Many issues are still to be standardized in the ATM adaptation layer. Available Bit Rate (AP) service is supposed to coexist with Variable Bit Rate (VBR) and Constant Bit Rate (CP). Congestion control through feedback to the source which specifies its maximum allowed cell rate is another issue. ATM switches are to send rate adjustment indication to the source (RM cell). Congestion should be signaled in forward direction. End to end flow control for Virtual Circuits (VC) should be specified.
User Network Interface (UNI) is going to support the third party call setup, implementation of group addresses, leaf initiated joins, and other features in the UNI 4.0.
Private Network-Network Interface (PNNI) will be defined to replace the Interim Interswitch Signalling Protocol (IISP). The available IISP specification supports SVC setup for small and static networks based on UNI signalling Q.2931 and has no routing or link state protocol specification. PNNI Phase I will supersede IISP.
In the area of ATM LANs, RFC 1483 addresses the multiprotocol encapsulation over ATM Adaptation Layer 5 (AAL5). It defines a Logical Link Control (LLC) encapsulation method which encapsulates the Protocol Data Units (PDUs) in an LLC/SNAP (Subnetwork Access Protocol) header. PDU's protocol is identified in the header. In this approach, multiple protocols are supported on one VC. In another method, each protocol has its own VC and as a result the VC identifies the protocol and no encapsulation is necessary. RFC 1577 addresses the classical IP and ARP over ATM. For all stations on a subnet that are connected to ATM, a Logical IP Subnet (LIS) is defined. Each IP address is mapped to a VC. The Inverse ARP and ATM ARP using an ARP server are also defined in this RFC. It only addresses the IP protocol. RFC 1680 describes considerations for IPng support for ATM services.
The LAN Emulation (LANE) specification by ATM Forum emulates LAN segments and appears as a MAC layer interface to bridges and layer-3 protocols (e.g. TCP/IP, IPX, AppleTalk). It supports broadcast and multicast. LAN segments and ATM endstations can be in the same Virtual LAN.
The Multiprotocol Over ATM (MPOA) is going to be specified. It separates the layer 3 from the physical layout of the network. It will define protocols for multiple LAN segments in a layer 3 subnet and multiple layer 3 subnets on the same interface. It will allow direct connections across layer 3 subnets. It will specify the protocols between layer 2/3 switches, route servers, and hosts. It will define a scalable network layer multicast and ATM Quality of Service (QoS) for layer-3 QoS.
Advances in computer hardware technology has resulted in faster and more powerful systems. At the same time the new applications with increasing demand for bandwidth and requirements for class of service are growing quickly. The bandwidth intensive transaction oriented applications, high traffic client/ server computing, multimedia support in applications with their real time and high bandwidth demands, imaging applications, all demand for more and more bandwidth and different classes of service.
Changes in the organizations, rightsizing, mobility and the nature of today's business operations dictate the need for on-going changes in the network.
ATM switches provide high and scalable backbone bandwidth, connectivity with low latency, and enable the internetworking to support bandwidth and delay sensitive applications. The logical formation of the network instead of reorganization of a network through physical changes reduces the cost of management. Virtual networking which implements a logical configuration for internetworking is the cost efficient solution for today and tomorrow.
The existing shared media LANs (Ethernet and Token Ring) will survive for a couple of more years. LAN switching is a solution that will gradually replace the existing congested traditional LANs. Eventually ATM-based switching will be the ultimate solution for internetworking.
[2] Brazdziunas, C., RFC 1680, "Ipng Support for ATM Services", Aug. 1994.
[3] Breault, R., McDyson, D., Editors, "ATM UNI Specification (V 3.0)", ATM Forum, October 1993.
[4] Chlamatac, I., Wong, E.W.M., "Interconnecting Multiple Input and Output LANs over a High Speed Gateway", IEEE International Conference on Communications, ICC, May 1993.
[5] Hadoung, T., et. al., "LAN, MAN, Frame Relay and ATM: A Taxonomy of Interworking Units", Proceedings of Interworking 1992, Nov. 1992.
[6] Halsall, F., "Data Communications, Computer Networks and Open Systems", Addison-Wesley, 1992.
[7] Heinanen, J., RFC 1483, "Multiprotocol Encapsulation over ATM Adaptation Layer 5", July 1993.
[8] IEEE Network Magazine, Special Issue on LAN Interconnection, Vol. 8, No. 5, Sep. 1991.
[9] IEEE Selected Areas Comm., Special Issue on Heterogeneous Computer Network Interconnection, Vol. 8, No. 1, Jan. 1990.
[10] Kavak, N., "LAN Interconnection over BISDN", Proceeding of Interworking 1992, Nov. 1992.
[11] Laubach, M., RFC 1577, "Classical IP and ARP over ATM", Jan. 1994.
[12] Onvural, R.O., "Asynchronous Transfer Mode Networks", Artech House, 1994.
[13] Onvural, R.O., Nilsson, A., Editors, "Local Area Network Interconnection", Plenum Press, 1993.
[14] Partridge, C., "Gigabit Networking", Addison-Wesley, 1994.
[15] Parulkar, G., "Local ATM Networks", IEEE Network Magazine, Vol. 7, No. 2, March 1993.
[16] Stassinopoulos, G.I., Venieris, I.S., "ATM Adaptation Layer Protocols for Signalling", Computer Networks ISDN Systems, Vol. 23, No. 4, 1992.
[17] Tanenbaum, A.S., Computer Networks, Prentice-Hall, 1989.
[18] Venieris, I.S., et. al., "Bridging Remote CL LANs/MANs Through CO ATM Networks", Vol. 15, No. 7, Sep. 1992.
He is a consultant with Neda Communications, Inc., serving the data communications and computer networking industry. He can be reached at kamran@neda.com.
