Advantages of Routing
One of the key themes that is developed throughout this chapter is the idea that routing is critical to scalable network design. Hopefully, this is not news to you. However, given the recent popularity and focus on extremely flat, “avoid-the-router” designs, a fair amount of attention is devoted to this subject. Many people are convinced that the key objective in campus network design is to eliminate as many routers as possible. On the contrary, my experience suggests that this is exactly the wrong aim—routers have a proven track record of being the key to achieving the requirements of campus design discussed in the previous section.
- Scalable bandwidth— Routers have traditionally been considered slower than other approaches used for data forwarding. However, because a routed network uses a very decentralized algorithm, higher aggregate rates can be achieved than with less intelligent and more centralized Layer 2 forwarding schemes. Combine this fact with newer hardware-based routers (Layer 3 switches) and routing can offer extraordinary forwarding performance.
- Broadcast filtering— One of the Achilles heels of Layer 2 switching is broadcast containment. Vendors introduced VLANs as a partial solution to this problem, but key issues remain. Not only do broadcasts rob critical bandwidth resources, they also starve out CPU resources. Techniques such as ISL and LANE NICs that allow servers to connect to multiple VLANs in an attempt to build flat networks with a minimal use of routers only make this situation much worse—now the server must process the broadcasts for 10 or 20 VLANs! On the other hand, the more intelligent forwarding algorithms used by Layer 3 devices allow broadcasts to be contained while still maintaining full connectivity.
- Superior multicast handling— Although progress is being made to improve multicast support for Layer 2 devices through schemes such as IGMP Snooping, CGMP, and 802.1p (see Chapter 13, “Multicast and Broadcast Services”), it is extremely unlikely that these efforts will ever provide the comprehensive set of features offered by Layer 3. By running Layer 3 multicast protocols such as PIM, routers always provide a vast improvement in multicast efficiency and scalability. Given the predictions for dramatic multicast growth, this performance will likely be critical to the future (or current) success of your network.
- Optimal path selection— Because of their sophisticated metrics and path determination algorithms, routing protocols offer much better path selection capabilities than Layer 2 switches. As discussed in the Spanning Tree chapters, Layer 2 devices can easily send traffic through many unnecessary bridge hops.
- Fast convergence— Not only do routing protocols pick optimal paths; they do it very quickly. Modern Layer 3 routing protocols generally converge in 5–10 seconds. On the other hand, Layer 2 Spanning-Tree Protocol (STP) convergence takes 30–50 seconds by default. Although it is possible to change the default STP timers and to make use of optimizations such as UplinkFast in certain topologies, it is very difficult to obtain the consistently speedy results offered by Layer 3 routing protocols.
- Load balancing— Routing protocols also have sophisticated load balancing mechanisms. Layer 3 load balancing is flexible, easy to configure, and supports many simultaneous paths. On the other hand, Layer 2 load balancing techniques such as the STP load balancing described in Chapter 7, “Advanced Spanning Tree,” can be extremely cumbersome and difficult to use.
- Flexible path selection— In addition to all of the other path selection benefits offered by routers, Cisco routers offer a wide variety of tools to manipulate path selections. Distribute lists, route maps, static routes, flexible metrics, and administrative distances are all examples of such mechanisms. These tools provide very granular control in a Layer 3 network.
- Summarized addressing— Layer 2 addresses use a flat address space. There is nothing about a MAC address that indicates physical location (it is much like a Social Security number). As a result, every bridging table in a flat network must contain an address for every node. On the other hand, Layer 3 addresses indicate location much like a ZIP code (postal code) or a telephone number’s area code. By allowing addresses to be summarized, this hierarchical approach can allow much larger networks to be built. As a result, forwarding tables not only shrink dramatically in size, the address learning or routing table update process becomes much easier. Finally, lookups in the forwarding tables can be much faster.
- Policy and access lists— Most Layer 2 switches have very limited, if any, filtering capabilities. When filtering or access lists are supported, they use MAC addresses, hardly an efficient way to implement policy. On the other hand, routers can be used to provide complex access lists that function on Layer 3 and 4 information. This is much more useful from a policy implementation perspective. Hardware-based access lists are becoming increasingly common and flexible in Layer 3 switches.
- Value-added features— Although it is unlikely that the switching router Layer 3 devices such as the Catalyst 8500 will support “high touch” WAN-oriented services such as DLSw+ and protocol translation, there are still a large number of extremely important features that are offered by these platforms. For example, technologies such as DHCP relay, proxy ARP, debug, and proxy GNS can be critical router-based features in campus networks. (Note that some Layer 3 platforms can perform “high touch” services by running them in software. For example, MLS using an RSM could do DLSw+ on the RSM. The native IP traffic uses the NFFC for wire-speed forwarding; the DLSw+ is dependent on slower software-based forwarding.)
Large networks almost always benefit from scalability, flexibility, and intelligence of routing. Try to build routing (Layer 3 switching) into your campus design.