UNIX System Administration Handbook - Evi Nemeth [221]
• A stand-alone network requires no routing.
• If there is only one way out of a network, clients (nongateway machines) on that network should have a static default route to the lone gateway. No other configuration is necessary, except on the gateway itself.
• A gateway with a small number of networks on one side and a gateway to “the world” on the other side can have explicit static routes pointing to the former and a default route to the latter. However, dynamic routing is advisable if there is more than one routing choice on both sides.
• If you use RIP and are concerned about the network and system load this entails, avoid using routed in active mode—it broadcasts everything it knows (correct or not) at short intervals. gated allows you to specify what routes may be sent out (“advertised”), thus reducing the flood of routing information. gated can also send RIP updates to particular gateways rather than broadcasting them everywhere.
• To have clients listen passively for routing updates without sending out their own information, use routed -q. Clients can also listen passively with gated, but gated has a larger footprint.
• Many people will tell you that RIP is a horrible, terrible protocol and that routed is the spawn of Satan. It isn’t necessarily so. If it works for you and you are happy with the performance, go ahead and use it. You get no points for spending time on an overengineered routing strategy.
• If RIP is not your primary routing protocol, you can have gated broadcast its routing information as RIP purely for the benefit of passive clients.
• routed listens to everyone and believes everything it hears. gated gives you more control over updates. Even if your site uses RIP, you may want to manage the exchange of routing data with gated and run routed only on client machines.
• Use dynamic routing at points where networks cross political or administrative boundaries.
• On dynamically routed networks that contain loops or redundant paths, use OSPF if possible.
• Routers connected to the Internet backbone that have a choice of upstream paths must use BGP. Most routers connected to the Internet have only one upstream path, and therefore a simple static route is sufficient.
A good routing strategy for a medium-sized site with a relatively stable local structure and a connection to someone else’s net is to use a combination of static and dynamic routing. Machines within the local structure that do not have a gateway to external networks can use static routing, forwarding all unknown packets to a default machine that understands the outside world and does dynamic routing.
A network that is too complicated to be managed with this scheme should rely on dynamic routing. Default static routes can still be used on leaf networks, but machines on networks with more than one router should run routed in passive mode. All machines with more than one network interface should run gated in active mode and broadcast routes with RIP.
14.8 CISCO ROUTERS
Routers made by Cisco Systems, Inc., are the de facto standard for Internet routing today. Having captured over 70% of the router market, Cisco’s products are well known, and staff that know how to operate them are relatively easy to find. Before Cisco, UNIX boxes with multiple network interfaces were often used as routers. Today, dedicated routers are the favored gear to put in datacom closets and above ceiling tiles where network cables come together. They’re cheaper, faster, and more secure than their UNIX counterparts.
Most of Cisco’s router products run an operating system called Cisco IOS, which is proprietary and unrelated to UNIX. Its command set is rather large; the full documentation set fills up about 4.5 feet of shelf space. We could never fully cover Cisco IOS here, but knowing a few basics can get you a long way.
IOS defines two levels of