Real World ATM
Articles and Tips:
01 Feb 1997
Editor's note: This article assumes that you have read the relatedarticle, "Understanding the Big ATM Picture," or that you already have a basic understandingof how ATM works.
After you understand the Asynchronous Transfer Mode (ATM) model and standards,you need to know how ATM is being implemented. Although an all-ATM networkbuilt on the User-to-Network Interface (UNI) and the Private Network-to-NetworkInterface (PNNI) is interesting in theory, in practice, few companies arejumping with joy at the prospect of throwing out existing Ethernet or Token-RingLANs and setting up an all-ATM network. Instead, most companies want tolink their traditional LANs to an ATM backbone or use ATM to create a high-speedworkgroup. To meet these needs, the ATM Forum and other organizations aredesigning ATM interoperability standards, and vendors are building productsthat adhere to these standards.
This article provides an overview of the following three ATM interoperabilityarchitectures: Classical IP Over ATM, LAN Emulation (LANE), and MultiProtocolOver ATM (MPOA). This article shows how these architectures enable you tointegrate ATM with traditional LANs and discusses each architecture's advantagesand disadvantages. In addition, this article explains how you can use IntranetWareto connect your network to an ATM backbone. (For more information, see "ATM in the IntranetWare Environment.")
CLASSICAL IP OVER ATM
Classical IP Over ATM was developed by the Internet Engineering TaskForce (IETF). This mature standard enables you to route IP packets overan ATM network, or cloud, using ATM as either a backbone technology or aworkgroup technology. Classical IP Over ATM maps IP network layer addressesto ATM addresses and enables ATM-attached devices to send IP packets overan ATM network.
How Classical IP Over ATM Works
Figure 1 shows the components of ClassicalIP Over ATM and how these components work together. In Figure 1, two end stations are connected to the ATM network. (An end stationcan be a workstation, a server, a bridge, a switch, or a router. TCP/IPdevices are called hosts or routers.) These end stations may be locatedin physically separate LANs, but because they are both connected to theATM network, they are part of the same logical subnetwork, or virtual subnetwork.
Figure 1: 1. End station A queries the ATMARP server for the ATM address of end station B. 2. End station A sets up a virtual circuit with end station B.
Each end station knows both its IP and ATM addresses and sends theseaddresses to an ATM address resolution protocol (ATMARP) server. The ATMARPserver maintains an address table and uses this table to translate betweenIP addresses and ATM addresses within a single virtual subnetwork.
An end station can use either permanent virtual circuit (PVC) or switchedvirtual circuit (SVC) connections. If an end station is using PVCs, theATM service provider or network administrator manually maps the IP addressesof reachable destination end stations to ATM virtual circuits. If a sendingend station is using SVCs, on the other hand, this station queries the ATMARPserver for the ATM address of the destination end station. The sending endstation then uses this address to set up an SVC.
After the PVC or SVC has been established, the sending end station convertsits packets into ATM Adaptation Layer (AAL) 5 cells and sends them overthe virtual circuit to the destination end station. The destination endstation then converts the AAL 5 cells back into IP packets.
How Classical IP Over ATM Stacks Up
Classical IP Over ATM has several disadvantages: Because the ATMARP servercan reach only one IP subnetwork, IP hosts can communicate directly onlywith destination IP hosts in that subnetwork. To send packets to a destinationIP host in another virtual subnetwork, a sending IP host must send the packetsthrough a router. The sending IP host uses a virtual circuit to connectto the router, and the router uses another virtual circuit to connect tothe destination IP host. Routers create a bottleneck because they are slowerthan switches.
Another disadvantage is that Classical IP Over ATM can route only IP.If you want to route other protocols such as IPX, you cannot use ClassicalIP Over ATM.
Furthermore, Classical IP Over ATM does not solve delay or congestionproblems because Classical IP Over ATM cannot take advantage of ATM's Qualitiesof Service. Finally, Classical IP Over ATM does not support multicast traffic(broadcasts sent to a specific group of hosts).
Classical IP Over ATM does have some advantages, however: You can obviouslyroute IP packets over ATM. Also, because end stations can be part of thesame virtual subnetwork even if they are on physically separate LANs, ClassicalIP Over ATM gives you a lot of flexibility in configuring your network.You can enable users who are hundreds of miles apart to share the same resourcesover a high-speed ATM backbone.
Developed by the ATM Forum, the LANE standard defines a way to bridgeLANs over a high-speed ATM backbone. LANE also enables you to connect workstationsrunning traditional protocols directly to an ATM network, creating a high-speedworkgroup.
With LANE, an end station on one LAN can communicate with an end stationon another LAN via an ATM network. An end station can also communicate withanother end station that is attached directly to the ATM network. As a result,you can include ATM-attached devices and non-ATM-attached devices on thesame network.
How LANE Works
End stations on different LAN segments can communicate with each otheras long as they are part of the same emulated LAN. Like a Classical IP OverATM virtual subnetwork, an emulated LAN is a logical LAN segment consistingof end stations connected by an ATM backbone. End stations on the same physicalsegment can be part of different emulated LANs, and end stations can bemembers of more than one emulated LAN.
LANE emulates the Media Access Control (MAC) sublayer of the Open SystemsInterconnection (OSI) model: LANE enables any network layer protocol thatworks with the OSI model (including IPX, IP, NetBIOS, and DECnet) to travelover an ATM network without modification. Users can then run applicationsover the ATM network just as they would over an Ethernet or Token-Ring LAN;the ATM network is invisible to LAN users.
The LANE standard consists of two components: LANE User-to-Network Interface(LUNI) and LANE Network-to-Network Interface (LNNI).
LUNI is defined in the LANE 1.0 standard and will be extended in theLANE 2.0 standard when it is finalized in March. LUNI specifies a way forend stations on an emulated LAN to communicate with each other over an ATMnetwork. An emulated LAN consists of the following components:
LANE clients, which are implemented on end stations
LANE services, which respond to requests from LANE clients
LANE services are provided by the following three components:
LANE server, which maps MAC addresses to ATM addresses
LANE configuration server, which gives LANE clients the ATM address of the appropriate LANE service
LANE Broadcast/Unknown Server (BUS), which transmits broadcast and multicast packets
All of the components are connected to the ATM network and to each other,usually using SVCs. (The LANE standard specifies that PVCs can be used,but most switch vendors do not support PVCs.) The LANE service componentscan also be implemented in the same switch. (Note: To avoid confusion, wedid not use commonly used acronyms, such as LEC, LES, and LECS, for LANEservices.)
Figure 2 shows how LANE works. Suppose thatyou were working on an emulated LAN and you wanted to access a file storedon a server that was located on a physically separate LAN. First, you wouldsend the file request, and your LANE client would determine if it knew theATM address of its LANE server. If your LANE client did not know this address,the client would query the LANE configuration server, asking for the ATMaddress of the LANE server.
Figure 2: 1. The LAN 1 end station queries the LANE configuration server for the ATM address of the LANE server. 2a. The end station queries the LANE server for the ATM address of the LANE client. 2b. If the LANE server does not know the ATM address, the LANE server queries the BUS. 3. The LAN 1 end station sets up a virtual circuit with the LAN 2 server.
After your LANE client received the correct address, your client wouldquery the LANE server for the ATM address of the LANE server on which thefile was stored. If the LANE server knew this address, the LANE server wouldsend the address to your LANE client. If the LANE server did not know thisaddress, the server would query the LANE BUS. The LANE BUS, in turn, wouldask all of the LANE clients on the emulated LAN for their ATM addresses.The LANE BUS would then return the correct address to the LANE server, whichwould return the address to your LANE client.
Finally, your LANE client would establish a virtual circuit to the serveron which the file was stored. Then the LANE client would convert its Ethernetor Token-Ring frames into cells and send these cells over the virtual circuitto the server.
With LANE, you can also send cells to other LANE clients without establishinga virtual circuit. Every LANE client that joins the emulated LAN establishesa permanent SVC with the LANE BUS. As a result, all LANE clients are connectedto the LANE BUS via SVCs. If you want to send cells to another LANE client,your LANE client would forward cells to the BUS via an existing SVC, andthe BUS would forward the cells to the destination LANE client. In thisway, LANE enables the connection-oriented ATM network to mimic a connectionlessnetwork.
LANE also allows you to broadcast cells to the entire emulated LAN. YourLANE client sends the cells to the LANE BUS, and the BUS forwards thesecells to all of the other LANE clients.
Because the LANE 1.0 standard does not specify a standard way for serversto communicate, each emulated LAN can have only one LANE server: Therefore,you cannot have redundant LANE servers, and the LANE server may become abottleneck in large emulated LANs. In addition, because the LANE servermust use virtual circuits to route traffic and because ATM switches cansupport only a limited number of virtual circuits (depending on your switches'capacity), using only one LANE server can overburden the switches.
The LANE 2.0 standard will solve this problem: LNNI defines a way forthe components of the LANE services to be distributed and defines the interfacebetween these components. As a result, you can have up to 20 LANE and BUSservers. (For information about Novell's upcoming LANE solution, see "What Is Novell Doing to Improve LANE?")
How LANE Stacks Up
LANE has several limitations. As with IP hosts on Classical IP Over ATMsubnetworks, an end station in one emulated LAN must use a router to communicatewith an end station in another emulated LAN. The sending end station mustestablish a virtual circuit with the router, which establishes a virtualcircuit with the destination end station. (See Figure 3.) Routers can create bottlenecks since routers are slower than switches.
Figure 3: To communicate with a LANE client on a different emulated LAN, LANE clients must use a router.
Because LANE is a bridging technology, it also suffers from scalabilityrestraints. LNNI addresses this problem by enabling you to have multipleLANE services. However, even LNNI limits you to a maximum of 2,000 end stationsper emulated LAN. In practice, the more end stations you have, the worseyour network will perform. More end stations mean more broadcasts and morevirtual circuits, which could be a problem since an ATM network can supportonly a certain number of virtual circuits at one time.
Moreover, LANE is an OSI data-link layer technology and is transparentto the higher layers of the OSI model. As a result, LANE cannot take advantageof ATM's Qualities of Service.
In addition, the LANE 1.0 standard supports only unspecified bit rate(UBR). The LANE 2.0 standard, on the other hand, enables you to specifythe type of traffic your emulated LAN will carry: constant bit rate (CBR),variable bit rate (VBR), UBR, or available bit rate (ABR). However, theLANE 2.0 standard requires all of the virtual circuits on your emulatedLAN to use the type of traffic you specify.
LANE does have several important advantages: For example, its abilityto route all OSI network layer protocols and run unmodified LAN applicationsover the ATM backbone makes LANE a powerful but simple way to take advantageof a high-speed ATM backbone.
In addition, because LANE enables end stations to transmit cells withoutestablishing virtual circuits, LANE can support connectionless LAN trafficbetter. LANE also supports broadcast and multicast traffic. Finally, LANE'sability to group end stations in an emulated LAN based on their MAC addressgives you flexibility in configuring your network.
MPOA was developed by the ATM Forum and should be finalized in April.MPOA enables you to route protocols such as IP, IPX, and NetBIOS from traditionalLANs over a switched ATM backbone. Like Classical IP Over ATM and LANE,MPOA provides OSI data-link layer bridging over a virtual subnetwork. Infact, MPOA incorporates LANE to provide bridging capabilities. Unlike ClassicalIP Over ATM and LANE, however, MPOA can route between virtual subnetworkswithout using traditional routers.
MPOA consists of the following:
Route Servers. Also known asMPOA servers, route servers maintain routing tables and calculate routes on behalf of edge devices. Route servers also communicate with traditional routers and with other route servers. Route servers are not necessarily one piece of hardware; their functions can also be built into existing routers and switches.
Edge Devices. Also known asMPOA clients, edge devices can be intelligent switches that forward packets and cells between LANs and ATM networks or network interface boards that forward packets and cells between ATM-attached devices and ATM networks.
Together, route servers and edge devices act as distributed routers:route servers determine where to send packets, and edge devices forwardthese packets.
How MPOA Works
Figure 4 shows how the MPOA components worktogether. LANs, workstations, servers, and routers are connected to edgedevices, which in turn are directly connected to the ATM network. As withClassical IP Over ATM and LANE, these components may be connected usingeither PVCs or SVCs.
Figure 4: 1. To perform cut-through routing, the edge device on subnetwork 1 queries the route server for the ATM address of the edge device on subnetwork 2. 2. The edge device on subnetwork 1 sets up a virtual circuit with the edge device on subnetwork 2.
If an end station on a LAN wanted to communicate with an ATM-attacheddevice, the end station would send a packet to the edge device. This devicewould check the destination MAC address or the network layer address ofthe packet. The device would then check its cache to see if it knew thecorresponding ATM address.
If the edge device did not know the ATM address, this device would querythe route server. If the route server knew the ATM address, this serverwould simply respond with the address.
If the route server did not know the ATM address, this server could useone of several routing protocols to communicate with other routers (bothtraditional routers and other route servers) to determine this address.These protocols include Routing Information Protocol (RIP), Open ShortestPath First (OSPF), Next Hop Routing Protocol (NHRP), and, when it is finalized,Integrated PNNI (IPNNI). (If you want to find out more about IPNNI, visittheNetWare ConnectionWorld-Wide Web [WWW] site at http://www.novell.com/feb.97/real27/ipnni27.html.)
When the edge device knew the ATM address, this device would establisha virtual circuit with the appropriate destination end station, convertits LAN packets into ATM cells, and send these cells to the destinationend station.
The edge device can establish a virtual circuit even if the destinationend station is on a different subnetwork; the edge device bypasses the routeserver when sending cells, sending them directly to the destination endstation. (See Figure 4.) This process is calledcut-through routingorone-hoprouting.According toLou Martinage, product marketing manager for Newbridge Networks, cut-throughrouting eliminates bottlenecks associated with traditional routers, lettingusers communicate at wire speeds no matter where their workstation is attachedto the network or in which subnetwork their workstation is located.
For short transmissions, however, cut-through routing may not be thebest approach because connection setup takes a long time relative to thesize of the transmission. Using a process calledhop-by-hop routing,MPOA can eliminate connection setup. With hop-by-hop routing, edge devicescan forward packets to the route server just as LANE clients can forwardpackets to the LANE BUS. Edge devices can also perform flow detection: Theycan forward packets to the route server, but if they detect a flow (a longtransmission), they can set up a virtual circuit to the destination endstation.
How MPOA Stacks Up
MPOA's biggest disadvantage is its relative newness. Although NewbridgeNetworks provides a prestandard MPOA solution, the ATM Forum has not yetfinalized the standard, and most vendors do not yet provide MPOA products.Depending on the implementation, MPOA can also add complexity to your network.
However, MPOA does provide many capabilities that neither Classical IPOver ATM nor LANE provides. Because MPOA is an OSI network layer technology,it has access to important network layer information such as traffic characteristicsand ATM's Qualities of Service. When establishing a connection, the edgedevice can use this network layer information to chart the best path toa destination end station based on the Qualities of Service the sendingend station requests.
MPOA also provides routing capabilities, which no other ATM interoperabilityarchitecture provides. With MPOA, you can route between traditional LANsconnected by a high-speed ATM backbone, thus creating a high-speed internetworkwithout the bottleneck of a traditional router. You can also use cut-throughand hop-by-hop routing to optimize both short and long transmissions.
WHERE DO YOU GO FROM HERE?
The standards explained in this article define how ATM works with traditionalLANs: Classical IP Over ATM allows you to integrate ATM with your IP networkas a workgroup or backbone technology. LANE enables you to integrate ATMwith your Ethernet or Token-Ring network as a workgroup or backbone technology.Finally, MPOA lets you connect LANs to an ATM backbone and route betweenthem, directly over the ATM network.
In addition to using standard architectures, vendors are creating proprietary,nonstandard architectures such as IP switching. If you are interested infind-ing out more about IP switching, visit theNetWare ConnectionWWWsite (http://novell.com/feb.97/real27/switch27.html).
As these standards and others like them are developed, ATM is maturing.To be sure, standards are appearing at a dizzying rate, and vendors arehaving a difficult time keeping up. However, vendors are keeping up, andusing the standards that exist today, they are building products that youcan use to connect traditional LANs to ATM workgroups and backbones. (Ifyou want more information about ATM products, visit theNetWare ConnectionWWW site at http://www.novell.com/feb.97/real27/vendor27.html.) As the development of ATM standards stabilizes, ATM may well become a commonlyused technology.
Kristin King works for Niche Associates, an agency that specializesin technical writing and editing.
ATM in the IntranetWare Environment
Michael Smith, director of Product Marketing for UB Networks
If your network is experiencing bandwidth crunch, you may be looking for a way to integrate Asynchronous Transfer Mode (ATM) with your network. To solve bandwidth bottlenecks on IntranetWare and NetWare networks, Novell has developed an easier way to implement ATM. IntranetWare supports ATM through its NetWare MultiProtocol Router (MPR) 3.1 compo-nent, which is a set of NetWare Loadable Modules (NLMs) that routes traffic over LAN and WAN media.
With IntranetWare, you can connect users on your LAN to a high-speed ATM backbone without installing ATM hard-ware on every workstation. You simply implement ATM on IntranetWare's NetWare MPR. (NetWare MPR 3.1 is also avail-able as a separate product for NetWare 4 and NetWare 3 networks.) To communicate with devices on a remote LAN, a workstation sends IPX, IP, or AppleTalk packets to NetWare MPR, which establishes a virtual circuit to send packets over the ATM backbone.
HOW NETWARE MPR WORKS
Figure 1 shows a network using NetWare MPR as an ATM connectivity device. The network consists of two LANs. Each LAN has a workstation and NetWare MPR, which connects to a high-speed ATM backbone using an ATM network interface board.
When a workstation on LAN 1 requests a file from LAN 2, the request passes to NetWare MPR, which sends the packets down through the multiprotocol router stack. (See Figure 2.) The packets pass to the IPX, IP, or AppleTalk layer, which adds protocol-specific information to the packets. The packets are sent to the Call Support Layer (CSL) if NetWare MPR must establish a virtual circuit to the destination workstation. If a virtual circuit has already been established, the packets are sent to the Link Support Layer (LSL). The CSL or LSL then sends the packets to a special Open Data-link Interface (ODI) layer for ATM, called ATM-ODI.
ATM-ODI includes three components:
ATM Wide Area Access module (ATMWAA)
ATM Topology-Specific Module (ATMTSM)
ATM Hardware-Specific Module (ATMHSM)
Within ATM-ODI, the packets first pass to the ATMWAA. If a virtual circuit has not been established, the ATMWAA tells the ATMTSM to set up a virtual circuit to the destination workstation.
The ATMWAA then sends the packets to the ATMTSM, which separates the connection-management functions (performed by the ATMWAA) from the hardware-specific functions (performed by the ATMHSM). If a virtual circuit has not been established, the ATMTSM instructs the ATMHSM to set up a virtual circuit. In addition, the ATMTSM breaks the packets down into ATM cells.
The ATMTSM then sends the packets to the ATMHSM, which is the ATM driver. The ATMHSM establishes a virtual circuit or uses a pre-established virtual circuit, as instructed by the ATMTSM. The packets pass to the ATM network interface board, which converts them to ATM cells and then to physical signals and sends them across the virtual circuit to NetWare MPR on LAN 2.
In NetWare MPR on LAN 2, the ATM cells pass through the ATM-ODI layer, where they are reassembled as IPX, IP, or AppleTalk packets. These packets then pass through the CSL or LSL layer and the IPX, IP, or AppleTalk layer and are sent to the server.
NETWARE MPR'S CAPABILITIES
NetWare MPR 3.1 supports several advanced ATM features. (You can take advantage of these features only if the ATM switch supports them as well.) NetWare MPR supports both permanent virtual circuits (PVCs), which are established when the network is configured, and switched virtual circuits (SVCs), which are established on demand. PVCs can reduce costs by using less expensive switches and less complex backbones. SVCs can save switch memory because they can be established and cleared as needed.
NetWare MPR 3.1 also supports both Logical Link Control (LLC) Encapsulation, which carries all protocols over the same virtual circuit, and virtual circuit multiplexing, which establishes a separate virtual circuit for each protocol. LLC Encapsulation reduces the number of circuits that must be established, whereas virtual circuit multiplexing enables you to establish different Quality of Service parameters for each protocol you are using (IPX, IP, or AppleTalk). By enabling you to use ATM as a backbone technology, by supporting both PVCs and SVCs, and by supporting both LLC Encapsulation and virtual circuit multiplexing, Novell's IntranetWare makes ATM flexible and easy to use.
What Is Novell Doing to Improve LANE?
Using a LAN Emulation (LANE) solution to integrate Asynchronous Transfer Mode (ATM) into your existing network can be somewhat complex. The LANE server, LANE configuration server, and LANE Broadcast/Unknown Server (BUS) are all included in most ATM switches, and they do their job transparently. However, you must still configure your workstations and servers as LANE clients, which communicate with LANE services.
Novell is developing a solution that will make the job easier by enabling vendors to include the LANE client in ATM-Open-Data Link (ODI) compliant drivers. The following vendors have ATM-ODI drivers that have been tested for functionality and stability: Efficient Networks, Inc., FORE Systems, Madge Networks, and ZeitNet, Inc.
To implement Novell's LANE solution, you must install the following components:
An IntranetWare, NetWare 4.11, or NetWare 4.1 server with an ATM network interface board and an ATM-ODI compliant driver
A workstation with an ATM network interface board, an ATM-ODI compliant driver, and a 32-bit client
An ATM switch
The LANE services from either ATM switches or routers
You will not need to replace your existing wiring because ATM runs on many types of cabling, including twisted pair, coaxial, and fiber. You will also not need to purchase new application software or modify your existing software because LANE works several layers below the application layer in the Open Systems Interconnection (OSI) model.
To find out more about Novell's LANE solution, call Novell at 1-800-NETWARE or 1-801-861-5588, or visit Novell's World-Wide Web (WWW) site. (Go to http://www.novell.com/intranetware. In the Table of Contents frame, choose Alphabetical List of Topics, and then choose Early Access Release. In the Early Access Release frame, scroll down to reach Novell ATM LAN Emulation.)
IPNNI: An ATM Routing Protocol
Integrated Private Network-to-Network Interface (IPNNI), which the ATMForum plans to release in 1998, is an extension to Private Network-to-NetworkInterface (PNNI). IPNNI will support traffic in a mixed IP and ATM networkby enabling IP and ATM routing devices to exchange routing information.With IPNNI, these routing devices will be able to route directly over theATM network.
HOW IPNNI WORKS
PNNI is a switching protocol that switches use to find the best paththrough an all-ATM network based on Quality of Service parameters, bandwidthcharacteristics, and delay constraints. IPNNI will provide the same capabilitiesas PNNI and will extend PNNI to act as a routing protocol that routers willbe able to use to find the best path through a mixed IP and ATM network.Traditional routers, ATM routers, and ATM devices that support IPNNI willall be able to use IPNNI to share information about the network topology.
For example, suppose that you had an ATM network connected to a traditionalLAN by an edge device (a device that bridges or routes traffic between theLAN and the ATM network), and suppose that IPNNI were running on this device.A router on the legacy LAN would send existing OSI network layer routingprotocols, such as Routing Information Protocol (RIP) or Open Shortest PathFirst (OSPF), to the edge device. This device would then encapsulate theseprotocols within PNNI and send them through the ATM network, using the bestroute that PNNI could find through the network. The edge device would thenreturn the routing information to the traditional router, which would calculatea path through the mixed IP and ATM network based on its understanding ofthe network topology.
IPNNI will be able to run on a traditional router, replacing RIP andOSPF, performing these protocols' routing functions, and using PNNI to determinethe ATM network topology and traffic characteristics. IPNNI will also beable to run on an ATM router. In either case, IPNNI will be able to communicatewith both traditional routers and other ATM routers.
It is important to understand, however, that IPNNI alone will not beable to integrate ATM with a traditional LAN. Instead, you must use IPNNIin conjunction with another ATM interoperability architecture, such as MultiProtocolOver ATM (MPOA).
Classical IP Over ATM, LAN Emulation (LANE), and MultiProtocol Over ATM(MPOA) are all standards-based architectures. IP switching, on the otherhand, is a proprietary architecture that was developed by Ipsilon NetworksInc. and explained in the informational Requests for Comments (RFCs) 1953,1954, and 1987. IP switching enables IP users to send IP packets transparentlyover an Asynchronous Transfer Mode (ATM) backbone.
HOW IP SWITCHING WORKS
Instead of using the ATM Forum's Private Network-to-Network Interface(PNNI) standards to forward cells, IP switching attaches a router (calledan IP switch controller) to an ATM switch, creating an IP switch. An IPswitch functions differently than a standard ATM switch functions: An IPswitch routes cells in a connectionless (hop-by-hop) fashion rather thanforwarding them in a connection-oriented fashion.
When IP packets are sent to an edge device (that bridges or routes trafficbetween a LAN and an ATM network) on an IP-switched network, the edge deviceconverts packets into cells and sends the cells to an ATM switch. The ATMswitch sends the cells to the router, which converts them into packets,determines where to send them, converts them back into cells, and sendsthe cells to the next switch. The switches and routers repeat this procedureuntil the packets reach their destination.
Because routing decisions are made by each switch rather than by thesource of the transmission, IP switching is a hop-by-hop, or connectionless,approach. Hop-by-hop switching has one advantage over ATM's connection-orientedswitching: IP switching has no connection setup and, therefore, no performancepenalty associated with connection setup. With small transmissions, theperformance penalty associated with connection setup can be relatively severe,so IP switching can benefit small transmissions.
Of course, there is still the performance penalty of routing cells ratherthan simply switching them--a penalty that impacts all transmissions, especiallylong transmissions. However, like MPOA, IP switching does provide flow detection,enabling the switches to detect long transmissions. Once the switch detectsa long transmission, the switch can bypass the router, forwarding cellsdirectly to the next switch. In this way, the switch no longer needs toroute cells.
HOW IP SWITCHING STACKS UP
IP switching has a few disadvantages: Because IP switching does not supportATM standards, it cannot take advantage of ATM's Qualities of Service. Also,at present, IP-switched networks can only route IP and protocols that canbe encapsulated within IP.
IP switching has some advantages, however: IP switching does enable youto send IP packets over an ATM network. In addition, because IP switchingis connectionless, it may support small transmissions better than a connection-orientednetwork would support small transmissions.
ATM Product Vendors
1-800-NET-3COM or 1-408-764-5000 http://www.3com.com
Bay Networks, Inc.
1-800-8BAYNET or 1-408-988-2400 http://www.baynetworks.com
Cisco Systems, Inc.
1-800-553-6387 or 1-408-526-7208 http://www.cisco.com
1-800-IBM-CALL or 1-404-238-1234 http://www.raleigh.ibm.com
Ipsilon Networks, Inc.
1-888-IPSILON or 1-408-990-2000 http://www.ipsilon.com
FORE Systems, Inc.
1-888-404-0444 or 1-412-635-3566 http://www.fore.com
1-800-876-2343 or 1-408-955-0700 http://www.madge.com
Newbridge Networks Corporation
1-800-DO-VIVID or 1-703-834-3600 http://www.vivid.newbridge.com
UB Networks, Inc.
1-800-777-4LAN or 1-408-496-0111 http://www.ub.com
1-800-999-9526 or 1-818-880-3500 http://www.xylan.com
* Originally published in Novell Connection Magazine
The origin of this information may be internal or external to Novell. While Novell makes all reasonable efforts to verify this information, Novell does not make explicit or implied claims to its validity.