The backbone of a network may be an architectural issue, but the network is the backbone of the organization. Keeping it running while upgrading and planning future requirements is some challenge. Network managers often leave their jobs for lighter work, such as nuclear physics.
Back in the 1970s, who would have thought every user would have a desktop computer? In the 1980s, who would have thought every desktop computer would eventually be connected to a network? Now it's the 1990s and the networks are already starting to weather. Time to plan for the 21st century.
Many agencies and departments have a need for an immediate increase in bandwidth and a future need for quality of service (QoS). Ever-increasing databases, multimedia requirements and intranets are devouring current bandwidth, while future videoconferencing and interactive dataconferencing require guaranteed fluid motion. That's QoS. Your mission is to figure out how to implement them both economically -- on shrinking budgets today and on questionable budgets tomorrow. What fun.
For some time, I have predicted that the whole networking infrastructure will change some time in the future. I think it ultimately will go back to dumb terminals and centralized processing with an optical fiber to everybody's desk. Network management will be a breeze. Needless to say, don't hold your breath, because nobody has figured this out yet. Of course, you could say "Freedman told me so," and maybe we could start a movement. Got a spare fiber?
Following are the major trends.
THE OSI MODEL
In order to understand networking, it is essential to become thoroughly familiar with the protocol stack in the OSI model. Once thought to become a worldwide communications system, the OSI model never came to pass, but still serves as an excellent teaching tool.
In a nutshell, the concept is that whatever is to be transmitted in the sending machine is passed from the application (top layer) to each layer of software below in the stack. Each layer adds its own header or encapsulates it in some manner and passes it on to the next one. At the bottom (layer 1), it goes out onto the cable. At the receiving machine, the processes start at the bottom and work their way back up the stack.
The layers of importance for this article are layers 2 and 3. Layer3 is the network layer, which adds the network address of the destination station. If the destination machine is not directly attached to the subnetwork or domain of the user, then the address of the closest router is used. Layer 3 is concerned with protocols such as IPX, IP, DECnet, AppleTalk and APPN. Closely associated with layer 3 is layer 4, which is the transport layer. For example, TCP is layer 4, and IP is layer 3.
Layer 4 is responsible for sequencing the packets and ensuring that everything sent is received properly. Layer 4 is implemented within the client and server, but unless there is a gross conversion of the protocol, layer 4 data is not dealt with by internetworking devices such as routers.
Layer 2 is the data link layer, also known as the MAC layer, which is actually Ethernet, Token Ring, FDDI and ATM (asynchronous transfer mode). Layer 2 provides the access method for creating the packets that are transmitted.
ROUTERS AND ROUTING
Routers have been an essential component in networking for more than a decade. Routers are used to convert traffic from LANs to WANs and vice versa, as well as to filter traffic within LANs for security and broadcast containment. Broadcasts occur when the address of a user or service is not known. Clients and servers may come online and announce themselves. Sometimes, servers continually announce themselves. In all cases, the broadcast has to reach all possible networks and stations that might be able to respond.
Routers inspect layer 3 of the message, which contains the network and station address of the packet's destination. If the packet is destined for the WAN, the router performs a layer 2 conversion by stripping off the LAN packet (Ethernet, Token Ring, FDDI), encapsulating the data into a WAN packet (Frame Relay, X.25, HDLC) and sending it on its merry way.
Because routers inspect layer 3 of the protocol, which has much more information than layer 2, they can determine the subnet (domain) the user is in and the type of message being sent (e-mail, Web access, file transfer, etc.). They can keep human resources traffic from getting to engineering and they can filter traffic based on "who should be doing what" kinds of criteria.
Routers have been used as network backbones for years. First as a series of routers cabled together, then as a collapsed backbone, using a type of router with a high-speed backplane similar to a switch. Standing at the center of the network, they improve performance by providing a central high-speed crossover between LAN segments.
The problem with routers is overhead. The common protocols in use (IPX, IP, etc.) are designed as connectionless systems, which means that the data message transmitted is split into multiple packets, each of which is a fully self-contained entity with source and destination addresses. Routers back up the protocol stack to analyze layer 3 of every packet they receive.
There is a trend toward multilayer switches, which can examine layer 3 and switch at layer 2. They inspect the first packet of the message (layer 3) and forward the remaining packets at layer 2 at higher speed because there is less processing at that layer. Cabletron has perhaps the most comprehensive multilayer approach, because it has implemented this on both LAN switches and ATM switches, while providing solid network management of the entire infrastructure. Ipsilon Networks provide a similar capability, but only for IP traffic over ATM (a bit more on this later).
There will always be a need for routing. The question is whether it is done in a router or in a device that does routing. Also, depending on which vendor you talk to, routers are either on the verge of extinction or thriving like mad. Cisco, being the largest router manufacturer, has a vested interest in high-end, high-priced routers much like IBM has in its mainframes. However, Cisco routers work well and many network managers are thoroughly familiar with them. IBM mainframes work well and many IS managers are thoroughly familiar with them. We could go on and on here.
VIRTUAL LANS AND VIRTUAL ROUTING
Traditionally, in order to improve traffic on a LAN that starts to get overloaded, a bridge or port configuration hub is used to split the LAN into multiple segments. These devices keep LAN packets transmitting within their own segment from wasting bandwidth in other segments. If a packet's destination is outside the segment, it is then bridged or routed to that segment.
LAN segments can be physically broken apart in the wiring closet, but virtual LANs, or VLANs, allow the operation to be performed logically rather than physically. Users are placed into a logical segment based upon their need to communicate with each other. Hence, traffic is contained within the logical segments, and users can be reassigned via software whenever they move to a different location.
Although the ATM Forum is expected to standardize VLANs on ATM switches in the future, there is no standardization of VLANs on LAN switches. Vendors' implementations are proprietary. In addition, although VLANs work at layer 2, they
still require a layer 3 router in order to send a message from one VLAN to another.
The problem with VLANs is that they are difficult to troubleshoot, because the logical segments don't mirror the physical segments. After all, that's the whole idea. They also don't scale well as networks expand and more routers are introduced. Managing VLANs, which deal with layer 2 MAC addresses in the midst of routers, which deal with layer 3 subnets and domains (IP, IPX, etc.), is a very complicated issue.
What the industry is working toward is virtual routing, which
lets the network manager view the entire enterprise network as a single routed entity.
FROM HUB TO SWITCH
There is an overwhelming move to gain immediate bandwidth by replacing Ethernet hubs with Ethernet switches. The standard 10BaseT Ethernet hub is a shared-media hub, which means that all users share the total 10Mbps bandwidth. The shared-media hub is actually a multiport repeater, because it sends any input signal to all output ports. It is also called an Ethernet concentrator in order to contrast it with the many highly-sophisticated, intelligent hubs on the market that do multiport repeating, switching, bridging, routing and brew your coffee at the same time.
Now, out goes the shared-media hub and in goes the switch. The switch has a backplane fast enough to cross over one port to another port at full-wire speed. If you have 12 ports on the switch, six pairs of ports can be connected simultaneously at wire speed. Instead of 12 users sharing 10Mbps, each user has a full 10Mbps to another user. For a couple of hundred dollars per port on average, we dramatically increase
Switched Token Ring provides the same advantage, as does switched FDDI. The best thing about implementing LAN switches -- also called frame switches -- is that you don't touch the user's machines or cables. That all stays intact while all the work is done in the wiring closet.
You can get a 10-fold increment by moving from 10BaseT to Fast Ethernet (100BaseT). You may also be able to use the same cables. 100BaseTX uses two pairs of Category 5 unshielded twisted pair, and 100BaseT4 uses four pairs of Category 3. Fast Ethernet is being implemented as a backbone running between 10BaseT Ethernet switches that have 100Mbps links. These switches multiplex the 10Mbps ports into the 100Mbps port. Another approach is wiring the 100Mbps link directly to the server.
For more dramatic speed, a Fast Ethernet switch can boost Fast Ethernet just as an Ethernet switch can boost regular Ethernet. But 100Mbps to a user's machine might be overkill, so instead of wiring each switch port to a single machine, the port can be wired to a shared-media hub that fans out to two or more users that don't need that much "oomph." This mixing and matching of switches and concentrators (hubs) lets you totally customize your bandwidth between clients and servers.
If you are upgrading PCs or purchasing new ones, think seriously about installing Ethernet 10/100 cards. For about $50 to $75 more than a standard 10BaseT card, you've got 100BaseT installed and ready for the future. New 10/100 cards provide auto sensing and can detect which topology is being used, so they can sit patiently waiting for their day.
ATM -- FUTURE BACKBONE OR BACK-BREAKER?
Perhaps nothing in the internetworking business has been hyped more than ATM over the past three years. Considering every major networking vendor has embraced it either by developing its own ATM products or by acquiring or aligning itself with an ATM company, it would appear that ATM is the future.
ATM is a switching technology that uses fixed-length cells that switch much faster than variable-length-packets. In addition, ATM is a connection-oriented system like the telephone. Once the call is set up, a pipe (virtual circuit) is set up from source to destination and the packets are pushed through without examining every last detail about them. Connectionless Ethernet and Token Ring systems, on the other hand, inspect every packet. ATM also provides quality of service, which means you can set up a guaranteed rate of delivery for realtime voice and video, and ATM is the only technology that is designed for both LANs and WANs. What's more, ATM is scalable from 25, 45, 100, 155, 622 Mbps and on into the stratosphere. What could be more ideal?
ATM has gained momentum as a way to provide a flexible, scalable LAN backbone that will boost the current infrastructure as well as support future needs. Almost every government agency and private corporation has planned for ATM or is at least thinking about it. But, it hasn't taken off as had been predicted. Integrating ATM into legacy LANs is not trivial. There's a difference in basic architecture between connection-oriented switches and connectionless LANs, which require some fancy footwork to blend together.
The ATM Forum has developed and is continuing to develop ways to do this, but the approaches are foreign to network managers accustomed to Ethernet and Token Ring, so many are playing a wait-and-see game. ATM sales in 1995 were 2.5 percent of all networking products, which seems small compared to the attention ATM has garnered in the industry. Pundits are already predicting the demise of ATM (a sure sign to buy more stock in ATM companies), while others feel ATM's glory days are still coming.
As you wade into the wild world of ATM, here are some of the technologies you will need to investigate.
LAN Emulation, or LANE, is an ATM Forum standard that supports legacy LAN packets within an ATM environment. A LANE driver runs in hardware called an edge device (also edge path adapter, network access device and LAN access device). The driver encapsulates Ethernet and Token Ring packets into LANE packets and then converts them into ATM cells. Nothing has to be done to the Ethernet and Token Ring stations, but native ATM clients also need the driver if they are to communicate with these legacy LANs.
(MULTIPROTOCOL OVER ATM)
MPOA is an upcoming standard from the ATM Forum that supports routing of legacy protocols (IPX, IP, etc.) over ATM. MPOA works at layer 3, whereas LANE works at level 2. MPOA separates the route calculations from the actual forwarding. When an edge device does not know how to forward a packet, it queries a centralized route server, which returns the destination address so that a switched virtual circuit (SVC) can be set up in the ATM switch. MPOA maintains the quality of service lacking in LAN Emulation and provides a virtual routing capability when fully implemented. Newbridge Networks is already shipping a preliminary version of MPOA, even though the standard isn't expected to be finalized until later this year.
I-PNNI is an ATM Forum standard that is fully supported by Bay Networks with Cisco also getting in on the act. I-PNNI is an extension of the PNNI (Private Network to Network) protocol, which ATM switches use to inform each other of the network topology so they can make appropriate forwarding decisions. With I-PNNI, a separate route server is not used, rather I-PNNI is implemented in edge devices and legacy routers, which can share information with the ATM switches. With I-PNNI, all devices have full knowledge of the network topology. Like MPOA, I-PNNI supports ATM's quality of service.
THE IP SWITCH
Ipsilon Network's IP switch is very controversial, because it offers exceptional performance, but is proprietary. The IP switch means exactly that: "switching IP." If you have a mixed shop, you will have to tunnel your other protocols within IP or change to IP to take advantage of it. A Pentium-based IP switch controller sits next to the ATM switch and routes short-duration traffic (address resolutions, SNMP, etc.) through the controller. Ipsilon claims an average of more than a million packets per second, trouncing the highest-performing routers that max out at a quarter of a million packets per second at their best.
(CELLS IN FRAMES)
Developed at Cornell University, CIF allows ATM backbones to be used with Ethernet LANs. CIF utilizes the inherent quality of service in ATM all the way to the Ethernet end station by placing the ATM cell within the Ethernet frame. It differs from LAN emulation, in which Ethernet packets are encapsulated into LAN emulation packets and then converted into ATM cells. CIF is implemented by replacing the Ethernet hub with a switch or multiplexor known as a CIF Attachment Device (CIF-AD). CIF drivers are used in the client stations.
It may seem like a daunting task, but understanding VLANs, frame switches, edge devices, route servers, multilayer switches, LAN emulation, MPOA, I-PNNI, IP switches and CIF are necessary to help you plan for the future. And, just when you get that under your belt, gigabit Ethernet will emerge to tantalize you with a high-speed backbone made from that good 'ole Ethernet cloth you know and love. Expected in 1997, Gigabit Ethernet could corner the market for LAN backbones, because articles like this make your head swim.
They may have a point. Will quality of service become unnecessary with a billion bits racing through the backbone? I wish I knew. I just spent the past three months researching this stuff, and let me tell you -- my brain has been racing at a billion bits per second.
In any event, good luck, and let me hear from you. I'll be studying nuclear physics in my spare time.
Following are some excellent white papers on the subject. While they tout the vendor's message, they are definitely worth reading.
Managing ATM Virtual LAN Workgroups
Newbridge Networks Inc., Herndon, VA Call 800/343-3600; 703/834-3600
Cabletron's Framework for Migrating to Virtual Enterprise Networking
Cabletron Systems, Rochester, NH
Call 800/332-9401; 603/332-9400
The Transition to ATM-based Switched Internetworking
Bay Networks, Santa Clara, CA
Call 800/8BAYNET; 408/988-2400
IP Switching: The Intelligence of Routing, the Performance of Switching
Ipsilon Networks, Palo Alto, CA
Call 888/IPSILON; 415/846-4600
Unified Communications over ATM at Cornell University
Alan Freedman's Computer Desktop Encyclopedia on CD-ROM is "the" award-winning reference about the computer industry. It contains more than 9,000 entries from micro to mainframe, including drawings, photographs, charts and specs on all important concepts, products and players.
Requires Windows 3.1, 95, NT
or Mac Softwindows. CD-ROM $39.95; Floppy $29.95. Site licenses start at $150. Call 215/297-5999 (FAX 8424). Look at examples and download the demo at .