by John Wobus  Upgrading Your Network Backbone What sort of upgrade should it be? Single unit versus centralized "cloud" versus distributed "cloud" Your backbone can consist of a single "box" at a single site, or might consist of a lot of units tied together with a LAN (such as an FDDI ring), with LAN links (full-duplex switched 100-Mbps Ethernet) or with ATM links. In the latter case, you might locate all the devices in a single place, bringing a piece of each of the outlying LANs to this central site, or you might have backbone equipment distributed in the buildings being served. The term "collapsed backbone" is used to refer to a backbone located at a single place, perhaps most often when it consists of a single device. The "single box" approach has some attraction: It is simple. Data traversing the backbone passes through just one device; arranging things so data passes through two or more devices i n series doesn't increase performance, offers more points of failure and adds to network latency. With a single central device, you aren't faced with the task of identifying which LANs are intercommunicating most often so you can reduce the number of hops between them: Every LAN is one hop away from every other LAN. You are not faced with the bandwidth constraints imposed by the LAN or links that interconnect backbone devices: Your one switch or router, for example, might have more internal capacity than an FDDI LAN. With a single device, you are more willing to pay money for features to make it fast or reliable, such as redundant power supplies. One disadvantage is that your device is a single point of failure for the whole backbone. However, if each of the distributed devices in a distributed backbone is just as likely to fail as a central switch or router, then to the client, the risk of not being able to reach the server across the backbone is actually less if there is only one such device to traverse. If the failure of a single device takes out everyone, the end user is not helped by the elimination of that possibility unless he/she has a way of using the working part of the backbone. For example, university students using open computer labs might simply have to go over to another lab if the portion of the backbone serving the first lab fails. On the other hand, full-time workers may not have much they can do if their own network connection is down even if other workers' network connections remain up. Another disadvantage is that the central device is likely to be unique in your network, so keeping a spare requires considerable additional expense. It is not always necessary, though, to have a complete spare: The device is likely to contain a number of identical interface cards, so you need keep just one spare card on hand. Another disadvantage is the possibility that that software, firmware or hardware changes will bring down your whole backbone. Again, this is not always true: many devices off er hot-swapping of interface cards and/or power supplies. Keeping the backbone at a central site, be it a single device or several interconnected devices, gives you maintenance advantages. In case of device failure, you have no worries about transporting devices to other buildings or getting access to the buildings. A single telephone at the central site gives you ready access to vendor support from the equipment location. You can locate test equipment and spare parts right where you need them. It also can provide you with a way to tap into any LAN attached to your backbone from a single room, which can aid in troubleshooting LANs. Also, a single backup power system can keep your whole backbone alive during power problems. There may be other safety measures that you can more likely afford at a single site than in sites scattered through buildings, such as fire prevention steps. The disadvantage is that you need to bring more interbuilding fiber into a single location. How much you distribute your network is very much of a trade-off of fiber versus electronics. You need to take into account that fiber represents a much longer-term investment than electronics: you are likely to get at least three times the length of usage from your fiber. Another disadvantage is that there always is some small risk of a disaster that you have not planned for, such as a building fire or flood. A distributed network can be designed to function even if one building were disabled in such a manner, where at a central location, your only counterbalancing advantage is that you can concentrate your preventative efforts at one place. One advantage cited for distribution of equipment is that data should not be brought to the central point if it is to stay at the "fringes" of your network If your centralized design imposes no bandwidth or latency restrictions on such data, and no appreciable risk or downtime is imposed, and the cost of the centralized backbone is no greater, then there is no disadvantage. Of course , for both the number of devices and the number of locations, there can be a middle ground. For example, a backbone consisting of just two devices has some of the advantages and disadvantages of a "single box" backbone as compared to a backbone consisting of 10 interconnected devices. Router versus frame switch versus ATM switch versus combination You may depend solely on routing, frame switching (Ethernet switching or FDDI switching), or ATM switching for your backbone irrespective of whether the network is centralized or not. In addition, you might have two elements in the backbone-for example, an ATM cloud and a single router, the commonly cited "one-armed router" configuration. Routing offers the greatest isolation between the LANs the backbone is interconnecting. This reduces the risk to the network as a whole when there are problems with one portion of it. This advantage is most attractive when there are departments that operate their LANs relatively independently from your central organization. Having the router can help contain problems like IP nodes stepping on each others' addresses, or faulty IPX servers claiming to be the clients' nearest server but offering no service. In a nonrouted network, very tight management and administration of the entire network, including manageable hubs, good records of port connections, good contact with the departments' system administrations, and a network management system that can track down MAC addresses, can help a central organization compensate for this lack of isolation. Routers also offer the opportunity to perform packet-filtering-based security. Although some switches offer this, they typically violate layering, using simple offsets rather than parsing headers, and thus can be defeated; or they shunt key packets to servers for parsing, usually increasing the time for "call setup." A problem with using packet-filtering in routers is that depending upon the router and the filtering, using the filtering features can reduce t he router's capacity appreciably. A natural advantage of routers is in handling WAN links: It is much easier to run a network that consists of 50 percent LAN and 50 percent T1 lines to remote sites using some routers. The well-known advantage of an ASIC-based switch is providing more performance for the price. They are generally simpler devices, and outside of VLAN and ATM features, likely to require fewer software or firmware upgrades. The flat network has some obvious administration advantages: IP addresses can be assigned out of one big pool and need not change when a desktop computer is moved from one location to another. The oft-quoted advantage of ATM switches is predicated on ATM's eventual takeover of networking: by moving into ATM now, you are getting over a conversion hump early. ATM has the potential for offering various qualities of service including low latency, which is touted as useful for real-time audio and video. It also has the potential to allow bandwidth to be reserved for various applications. It also can be used for WAN links, allowing you to treat your WAN almost like a LAN, and it handles trunking (using multiple links between two switches to provide more bandwidth between them) and the upgrading of links in a standard and straightforward manner. ATM also offers ELANs, the only standard version of tagged VLANs. Furthermore, ATM may be the heart of future PBX technology as well as organizational CATV-like video services. The disadvantage of ATM is that it is very young and very complex. A lot of it is still in development, including many of the features that are supposed to make conversion to it from a frame-based environment smoother: by jumping early, you forgo these advantages. The protocols being used in ATM (PNNI, LANE and MPOA) are very complex and still in development, suggesting a future of software and firmware upgrades to correct the inevitable problems with implementations of such complex protocols, and to bring devices up to the newer versions of the protocols. All this complexity is handled by software in servers, and all of it provides services only up to Layer 2: the service provided by Ethernet NICs is provided in software by a number of servers, with the added complexity necessary to allow the servers to back each other up. In general, this complexity is isolated from affecting a data transfer in progress, but it does affect the call setups-lengthening them, limiting their number, and increasing the risk of problems. The administration of ATM networks reflects this complexity: another layer of administration is necessary. Also, ATM's youth means its management and troubleshooting tools are also young. Some of the touted features are not there yet-quality of service for everyday LAN applications, for example. And finally, we can not be certain what its future will be: it could become as ubiquitous as Ethernet and TCP/IP, or its development may proceed more like that of some other well-known standardized network technology efforts: FDDI, Token-Ring, OSI and ISDN, for example. The combination of router and switch is an oft-touted choice now. A site using routers moves to a "one-armed router" configuration with a switch or a number of switches that offer VLANs (or ELANs) with a router to carry traffic between the VLANs. This is a fairly natural step for a site trying to move away from routers toward a flat network or toward VLANs or ATM. On day one, before any administrational changes are made to combine or reorganize the subnet structure, the router is still doing the same work it would before, so the switching configuration is nothing more than a very expensive plug-board. If your goal is to flatten, or reduce the number of subnets, or to start combining them in location-independent ways or make use of to-the-port VLANs, then the configuration allows you to proceed with the administration involved in this process. If you have central servers that serve groups of clients in one or a few locations, then even without to-the-port VLANs, you can utilize VLAN-based subnets that put clients on the same subnet as their server. Next Upgrading Your Network Backbone Is It Time To Upgrade? What Do You Need To Know? Evaluating Proposals Additional Issues Updated March 14, 1997 Print This Page E-mail this URL