There are many ways in which multiple routing protocols can be implemented on a network. Sometimes, the protocols aren't even aware of each other (referred to as "ships in the night") and exchange no information. This is often the case during complex migrations. The other end of the spectrum is a fully-dynamic relationship, such as when "mutual-redistribution" is configured. Somewhere in the middle is "best practice".
In my last tip, I discussed using Interior and Exterior Gateway Protocols (IGP and EGP) together, and in this tip, I want to focus on the interaction of these protocols and go into some more detail on the ways you should go about implementing them.
To start, you need to clearly understand what you hope to accomplish, because these same two sets of protocols are used in two very different ways. This distinction revolves around whether or not your network is going to be a transit network. That is, do you plan to allow packets to pass through where your network is neither the source nor destination? If you're an ISP, then the answer is likely "yes" and if you're a regular enterprise, the answer is usually "no".
The answer to that question guides your design principles for manageability. Specifically, the IP subnets advertised through transit networks are often very dynamic and it's impractical to statically configure each IP route. Enterprise networks, on the other hand, may make a lot of changes internally, but their address ranges don't change very often.
To the enterprise network administrator, this choice often boils down to whether it is easier to configure mutual redistribution between their IGP and EGP, and control the networks with access-lists, or whether it is easier to configure both the IGP and EGP with network statements. Often, network statements lose this battle because they are perceived as static, which is true enough, but the access-lists that drive route-maps are also static and thus mutual redistribution is no longer dynamic. Mutual redistribution is almost never a good idea in an enterprise network.
Network statements, while the commands themselves are static, are simple to understand and result in very deterministic behavior. No need to guess what your network is going to do. But many network admins don't realize network statements are also somewhat dynamic, because even though you configure the statement, that route won't be advertised unless it exists in the routing table already (i.e. the router has an IGP path, or static route or connected interface), so if the IGP route goes away, so does the BGP route.
Even so, another popular option is one-way redistribution. This is useful when you have multiple paths to exit the network and you need a good balance between dynamic and controlled.
Finally, remember that using two protocols gives you the flexibility of intelligent decision-making in the EGP without sending 120,000 routes into your IGP. Filter as much as possible, ideally everything but the default.
Tom Lancaster, CCIE# 8829 CNX# 1105, is a consultant with 15 years experience in the networking industry, and co-author of several books on networking, most recently, CCSPTM: Secure PIX and Secure VPN Study Guide published by Sybex.