Should the Flat Earth Model Be Used?
Before running off to work on the design, a meeting was arranged with the key network personnel from Happy Homes. Knowing that many people associate switching with very flat networks, the design team wanted to get a feel for the client’s expectations about the overall design.
Several members of the Happy team went to the white board and started drawing the network they envisioned. Their final drawing is reproduced in Figure 17-2.
Figure 17-2. The Design Envisioned by the Happy Homes Staff
As the Happy staff described their design, it was clear that they subscribed to the campus-wide VLANs model discussed in Chapter 14, “Campus Design Models.” Some of the key features they mentioned are included in the following list:
- Using 30 or 40 VLANs to create lots of communities of interest. This would provide fine-grained control over security and broadcast radiation.
- Because every VLAN would be configured on every switch, it would be easy to place users in different buildings and floors in the same subnet. This would allow traffic within the team to avoid the “slowness of routing.”
- It would be easy to add a new user to any VLAN. A vendor had recently demonstrated a drag-and-drop VLAN assignment tool that had the whole team very excited.
- An ATM core could be used to trunk every VLAN between every building. The ATM core could also provide multiservice voice and video capabilities in the future.
- Servers could be attached to multiple VLANs/ELANs with LANE and ISL NICs. Again, this would provide a direct Layer 2 path from every user to every server and minimize the use of routers.
- VMPS could be used to dynamically assign VLANs to the rapidly growing number of laptop computer users. This would allow VLANs to follow the users as they moved between various offices and conference rooms. Their Cisco Sales Representative had also demonstrated a product called User Registration Tool (URT). This would allow VLAN assignments to be made based on NT Domain Controller authentication. The flexibility of this feature excited the Happy Homes network staff.
- The small amount of remaining inter-VLAN traffic could be handled by a pair of 7507 routers. These routers would use one HSRP group per VLAN to provide fault-tolerant routing for each VLAN.
The design crew had run into these sorts of expectations in the past. In fact, they had recommended almost this exact design to several clients one or two years earlier. At that time, the design team felt that avoid-the-router designs were necessary because software-based routers could not keep pace with the dramatic growth in campus bandwidth demands.
However, the results of these campus-wide VLANs created many unforeseen problems, including the following:
- Spanning Tree loops and instability were common. These frequent outages were often campus-wide and difficult to troubleshoot.
- Even when fast-converging protocols such as EIGRP, HSRP, and SSRP provided 5–10 second failover performance, Spanning Tree still created 30–50 second outages.
- Drag-and-drop VLAN assignment schemes were never as easy to use as everyone expected. Not only were some of the management platforms buggy, the VLANs-everywhere approach made it almost impossible to troubleshoot network problems. Instead of jumping in to solve the problem, network staff found themselves spending lots of time simply trying to comprehend the constantly changing VLAN layout.
- The performance of ISL and ATM NICs was disappointing and also led to unexplainable server behavior.
- It was becoming harder and harder to keep traffic within a single VLAN, the very premise of flat earth networks.
On the plus side, the design team did acknowledge that in certain situations, the advantages of the campus-wide VLAN model might outweigh its downsides. For example, the design team had recently worked on a large design for a university. The school wanted to create separate VLANs for students, professors, and administrative staff. Furthermore, they wanted these communities to be separate across each of the university’s departments (for example, the biology department and physics department would use six VLANs: two for administrative staff, two for students, and two for professors).
Because the school assigned a laptop to every student and professor, it was impossible to make static assignments to these VLANs. Campus-wide VLANs and URT/VMPS allowed students and university personnel to simply plug into any available outlet and receive the same connectivity throughout the entire campus. However, the inter-VLAN routing could still be centralized, allowing for much simpler access list configuration.
The design team mentioned that they also had some large hospital and government clients using similar designs. However, because the design team did not see the need for this sort of dynamic VLAN assignment and centralized access lists in the Happy Homes network, they recommended against this approach.
The discussion continued throughout the day. During this time, the design team brought up other issues discussed in Chapter 7, “Advanced Spanning Tree,” Chapter 11, “Layer 3 Switching,” Chapter 14, “Campus Design Models,” and Chapter 15, “Campus Design Implementation.” In the end, Happy Homes decided that the risks of the campus-wide VLAN approach were too great. They agreed that Layer 3 switching eliminated virtually all of the downsides they associated with traditional routers. Therefore, rather than striving to avoid Layer 3 routing, they decided to use it as a means to achieve stability, scalability, simplicity, and ease of management.
As the design team left at the end of the day, they agreed to return in several weeks with two separate designs for Happy Homes’ consideration. Although both designs would utilize Layer 3 switching, one would blend Layer 2 and Layer 3 processing and the other would maximize the Layer 3 component. The results of their design efforts are presented in the following two sections.