White Paper: Routers/WAN access devices white paper

Combining elements of routers, hubs and switching in one device can reduce LAN complexity and cut costs

Combining elements of routers, hubs and switching in one device can reduce LAN complexity and cut costs

Mark Twain once said: "Put all your eggs in the one basket and, watch that basket.'' That, in essence, is what many companies are doing with their networks. They are moving to a networking architecture called a collapsed backbone in which all LANs are attached to one device that passes traffic from one LAN segment to another. Typically, the device used in the center of a collapsed-backbone network has been a router, but today it might be a hub or an ATM (Asynchronous Transfer Mode) switch.

Collapsed-backbone networks are fundamentally different from traditional backbone networks in which each LAN is connected via a router or a hub to a backbone cable or fiber that runs throughout a building. In contrast, a collapsed-backbone network connects users to hubs that, in turn, are connected to a centrally located, high-performance router.

Using a collapsed-backbone network offers many benefits over using distributed networks. First, complexity is centralized, which makes the network easier to manage. Rather than dispersing routers throughout an organization, you can put a single router near the technical-support staff.

Compared to hubs, routers are complex to install, configure, and manage. This means that in a distributed network, networking staff are frequently dispatched to locations throughout a company's facilities to maintain the routers. With a collapsed-backbone network, however, you can keep a router in one spot and install hubs throughout the organization. This saves the recurring labor cost of sending a technician to every floor of a building whenever a problem with a device arises.

Adopting a collapsed-backbone architecture also provides an economy of scale. Installing one large router with inexpensive hubs is less expensive than providing many small routers for every floor or department.

Another reason to migrate to collapsed-backbone networks is that they provide the centralized management benefits of the old IBM mainframe environments. For example, a company can consolidate diagnostic and troubleshooting equipment in one location; as a result, the company requires less equipment than it would with a distributed-network architecture.

Collapsed-backbone networks also can provide a higher degree of security than distributed networks. For instance, access to a collapsed-back-bone router is frequently controlled in much the same way as access to data centers was controlled in the days of mainframes. The central router typically is placed in a room to which access is restricted. In contrast, a distributed network, in which access to routers is dispersed throughout a corporation, may be harder to control.

The one drawback of a collapsed-backbone approach is that it introduces a single point of failure in the network. Equipment manufacturers however, have addressed this problem by building into router and hub chassis such features as redundant cooling fans and power supplies, as well as by using modular components that can be swapped in and out without bringing down the network.

Although these features do not make the device completely fault-tolerant, they do reduce downtime. Often, the only point of failure is a router's backplane, over which all the LAN-to-LAN traffic must pass. If that fails, the network goes down. But the same problem plagues distributed networks when backbone cabling is damaged. And the chance of cable damage occurring is greater, because the cabling runs throughout a building and is susceptible to accidental cuts.

A changing landscape

Collapsed-backbone networks have long been the domain of high-end router vendors such as Cisco Systems (Menlo Park, CA), Proteon (Westborough, MA), 3Com (Santa Clara, CA), and Wellfleet Communications (Billerica, MA). These vendors have long offered high-performance routers with the capacity (both in backplane bandwidth and packet-processing power) to handle the large volume of traffic that must pass between LANs in a collapsed-backbone environment.

Routers used in distributed-backbone networks do not have such high performance requirements because the bulk of backbone traffic passes over cabling. And the backplanes of departmental routers (i.e., those attached to the backbone cabling) carry only the packets destined for the LANs attached to that one router.

The collapsed-backbone landscape is changing, however. High-end hubs from such vendors as Cabletron Systems (Rochester, NH), Chipcom (Southborough, MA), IBM (White Plains, NY), Lannet (Irvine, CA), Standard Microsystems Corp. (Hauppauge, NY), Synoptics Communications (Santa Clara, CA) and 3Com now employ switching and bridge and routing modules for use in collapsed-backbone networks. A collapsed-backbone network can also be built using ATM switches, such as those offered by Fore Systems (Pittsburgh, PA) and Ungermann-Bass (Santa Clara, CA).

When designing a collapsed-backbone network, you must weigh several factors before choosing among a router, a hub, and an ATM switch. Considerations include the architecture of the existing network, the type and amount of traffic on the network and whether the delivery of the data is time-sensitive. Your decision may also be influenced by corporate networking philosophy. Some companies, for example, design their networks around enterprise hubs. Because they have expertise in that product area (and a large investment in the hub's management system), they may stay with hubs rather than moving to a network based on stand-alone routers or ATM switches.

The overall networking environment determines which technology - routing, hubs or ATM switches - you should use. "It's not a matter of hub-based switching versus routing,'' notes Chris Bennett, a product manager at 3Com. "Each technology lends itself to specific environments. Routers, for instance, are well suited to handling a mix of network types and protocols. It's quite common to see a router-based collapsed-backbone network with Ethernet, token-ring and FDDI (Fiber Distributed Data Interface) LANs all connected to a single router. In hub-based collapsed-backbone networks, all LAN segments are typically of the same type."

Routers are also ideally suited to networks that require advanced traffic filtering. Because routers operate at layer 3 - the networking layer - of the OSI (Open Systems Interconnection) model, they can offer more sophisticated traffic filtering than other internetworking devices that work at lower layers. From a practical standpoint, router filtering allows network managers to set up what are commonly called firewalls, which keep traffic confined to a LAN segment.

Such firewalls are used in two ways. First, they can keep unwanted traffic from flooding a network. For instance, an application used by one department may employ a chatty networking protocol in which many exploratory or broadcast packets are sent onto the network. You can use a router to filter that traffic and confine it to that department's LAN. Doing this prevents the traffic from spreading over the entire corporate network, where it can consume large amounts of bandwidth and degrade network performance.

Firewalls also are commonly used to maintain security, because you can configure a router so that users on one LAN segment cannot access another network. A router's filtering capabilities can help in several ways. Filtering confines packets destined for users on the same LAN to that LAN, preventing a person on another LAN segment from eavesdropping on these packets. And with filtering, you can deny a user access to the network resources on a particular LAN segment. For instance, you can keep users away from file servers on LAN segments that they are not authorized to access. This helps maintain the confidentiality of employee records such as salary information and reviews.

The downside to routers is their complexity. They have a reputation for being hard to configure, a difficulty that hub vendors have tried to capitalize on. Typically, hubs are easier to maintain and provide certain management functions not commonly available with routers. However, hubs are not suitable for all networking environments. For instance, they cannot perform the type of advanced filtering available with routers. And because they do not translate packet formats from, say, FDDI to Ethernet, they are not as adept at handling very mixed-type networks.

Hubs are suited to networks in which most LANs are of one type such as Ethernet. But even in this case, hubs have not typically been used as the lone central device for a collapsed-backbone network. In the past, hubs, even large chassis-based enterprise hubs with backplanes designed to handle large volumes of LAN traffic, still required a high-performance router (either a stand-alone unit or one plugged into the hub chassis) to create a collapsed backbone. In such a configuration, the different LAN segments were interconnected by passing the traffic over the backplane of the router, not through the hub.

This situation has changed as hub vendors have integrated switching into their hubs. Hub vendors Cabletron, Networth (Irving, TX), Optical Data Systems (Richardson, TX), SMC, Synoptics and 3Com now offer Ethernet switching products. Ethernet switching pioneers Alantec (San Jose, CA), Kalpana (Sunnyvale, CA), and Lannet have beefed up their offerings for the collapsed-backbone market.

All these vendors offer products that let you use high-end hubs as the center of a collapsed backbone. They use Ethernet switching modules that couple multiple LAN segments through a high-speed switching matrix. Typically, the switching matrix is capable of sustaining multiple, simultaneous connection paths between LAN segments. Ethernet switching lets you dedicate a full 10 Mbit/s to a LAN segment or even to a single workstation, instead of having all users and LAN segments share a single 10Mbit/s pipe.

In contrast to the way that a router handles traffic, Ethernet switching hubs switch LAN traffic on a packet-by-packet basis, using address information contained within each packet's layer 2 (the MAC [media access control] layer) rather than layer 3 (the network layer), as a router would. In essence, that means Ethernet switching hubs act as multiport bridges, which unfortunately also means that these hubs are unable to perform the advanced filtering of a router.

However, backbones built around an enterprise hub with integrated Ethernet switching are easier to configure and manage. For example, hubs with Ethernet switching automatically know the MAC address of all devices attached to it. "This gives me a tremendous advantage, [because] the new product integrates easily into my network and the learning curve is al most zero,'' says Al Herrington, communications manager at St. Jude Children's Research Hospital (Memphis, TN) and a beta user of the ONcore Switching System, a new switching hub from Chipcom.

Separate realities

The advantage to knowing the MAC address of each workstation on a network is that it eases one of the most common management tasks: the handling of moves, additions and changes. Hubs have always been able to handle the changes that occur when a user moves from one location within a company to another.

Switching, however, introduces a new twist. Traditionally, handling changes meant dealing with users who had physically moved; now it goes beyond that.

Location used to determine who was on a LAN; all the users in one department were in the same location and on the same LAN. Now, users' current projects often determine which group they are assigned to.

This creates a new type of networking architecture called virtual LANs, in which users are connected not according to location but according to logical requirements. For instance, a product development team may include a design engineer, a marketing person, someone from accounting, and a member of upper management. These people may be scattered over an organization, but they need access to each other and to common information.

One way to connect them is to rewire the building so that each person's workstation connects to a single hub for the group. But this approach is usually impractical. Instead, using the management features of a switching hub, you can assign project members to a virtual workgroup. For instance, the management system for Lannet's LET series of hubs allows a network administrator at a management console to use a mouse to "tag" a user and then drag and drop that user into a logical LAN segment.

This capability not only helps build workgroups based on work projects, but it also lets an administrator break large, congested Ethernet groups into smaller segments. This "pushes bandwidth out to the user,'' says Jim Goede, a product manager at Lannet. "With switching, you can dedicate a full 10 Mbit/s of bandwidth to a smaller number of users or even to a single user,'' he adds. With some applications, such as providing compressed video to the desktop, the ability to supply a dedicated 10 Mbit/s of bandwidth to a small number of users means the difference between an application running on the network and its refusal to run.

Such segmentation is also possible with an ATM switch. In fact, most companies use ATM technology to connect collapsed-backbone routers or hubs in many buildings. However, in situations in which users require more bandwidth, or when time-sensitive traffic is running on the network, an ATM switch can be used as the center of a collapsed backbone.

Several vendors, including Fore Systems and Ungermann-Bass offer ATM switches that are designed with collapsed backbones in mind. The benefits of using ATM rather than router and Ethernet switching-hub technologies are scalability and its connection-oriented service. ATM can deliver more bandwidth to a LAN or a single user than Ethernet or FDDI can.

Also, the connection offered by ATM is inherently less delay-sensitive than those offered by the other technologies. With ATM, when two workstations want to communicate they set up a session in much the same way a phone call is set up. Once the session is established, the data stream passes from one node to another. That can be useful in situations in which time sensitivity is important, such as in a multimedia application that must synchronize the delivery of moving images, data and graphics files, and voice traffic.

Future considerations

Sometimes elements of all three technologies may be necessary to build a backbone network. Many vendors recognize this fact and have taken steps, through acquisitions or mergers, to acquire the technology they lack. This summer, Synoptics, a leading enterprise-hub vendor, and Wellfleet, a leading supplier of collapsed-backbone routers, announced that they would merge. The resulting company will have the Ethernet switching, ATM and routing technology for any type of collapsed-backbone network.

Many hub and router companies have acquired companies with switching technology. For instance, Cisco Systems purchased switching-hub vendor Crescendo Communications (Sunnyvale, CA). Network Systems (Minneapolis, MN) acquired Bytex. 3Com acquired Synernetics and Chipcom merged with switching-hub vendor Artel Communications (Hudson, MA).

But simply having the technology is not necessarily enough; these companies must now put it to good use. Fortunately, internetworking firms are starting to formulate long-term strategies for incorporating the elements of the three technologies into a single network. This is a crucial move for any company that pegs its network future on a collapsed-backbone architecture.

( Byte 1999

Compiled by Will Garside


Read more on Networking hardware