Hierarchical Network Design Principles

Hierarchical Network Design Principles

Just because a network seems to have a hierarchical design does not mean that the network is well designed. These simple guidelines will help you differentiate between well-designed and poorly designed hierarchical networks. This section is not intended to provide you with all the skills and knowledge you need to design a hierarchical network, but it offers you an opportunity to begin to practice your skills by transforming a flat network topology into a hierarchical network topology.

Network Diameter

When designing a hierarchical network topology, the first thing to consider is network diameter. Diameter is usually a measure of distance, but in this case, we are using the term to measure the number of devices. Network diameter is the number of devices that a packet has to cross before it reaches its destination. Keeping the network diameter low ensures low and predictable latency between devices.

Rollover Network Diameter button in the figure.

In the figure, PC1 communicates with PC3. There could be up to six interconnected switches between PC1 and PC3. In this case, the network diameter is 6. Each switch in the path introduces some degree of latency. Network device latency is the time spent by a device as it processes a packet or frame. Each switch has to determine the destination MAC address of the frame, check its MAC address table, and forward the frame out the appropriate port. Even though that entire process happens in a fraction of a second, the time adds up when the frame has to cross many switches.

In the three-layer hierarchical model, Layer 2 segmentation at the distribution layer practically eliminates network diameter as an issue. In a hierarchical network, network diameter is always going to be a predictable number of hops between the source and destination devices.

Bandwidth Aggregation

Each layer in the hierarchical network model is a possible candidate for bandwidth aggregation. Bandwidth aggregation is the practice of considering the specific bandwidth requirements of each part of the hierarchy. After bandwidth requirements of the network are known, links between specific switches can be aggregated, which is called link aggregation. Link aggregation allows multiple switch port links to be combined so as to achieve higher throughput between switches. Cisco has a proprietary link aggregation technology called EtherChannel, which allows multiple Ethernet links to be consolidated. A discussion of EtherChannel is beyond the scope of this course. To learn more, visit:

http://www.cisco.com/en/US/tech/tk389/tk213/tsd_technology_support_protocol_home.html

in figure, computers PC1 and PC3 require a significant amount of bandwidth because they are used for developing weather simulations. The network manager has determined that the access layer switches S1, S3, and S5 require increased bandwidth. Following up the hierarchy, these access layer switches connect to the distribution switches D1, D2, and D4. The distribution switches connect to core layer switches C1 and C2. Notice how specific links on specific ports in each switch are aggregated. In this way, increased bandwidth is provided for in a targeted, specific part of the network. Note that in this figure, aggregated links are indicated by two dotted lines with an oval tying them together. In other figures, aggregated links are represented by a single, dotted line with an oval.

Redundancy

Redundancy is one part of creating a highly available network. Redundancy can be provided in a number of ways. For example, you can double up the network connections between devices, or you can double the devices themselves. This chapter explores how to employ redundant network paths between switches. A discussion on doubling up network devices and employing special network protocols to ensure high availability is beyond the scope of this course. For an interesting discussion on high availability, visit:

http://www.cisco.com/en/US/products/ps6550/products_ios_technology_home.html.

Implementing redundant links can be expensive. Imagine if every switch in each layer of the network hierarchy had a connection to every switch at the next layer. It is unlikely that you will be able to implement redundancy at the access layer because of the cost and limited features in the end devices, but you can build redundancy into the distribution and core layers of the network.

In the figure, redundant links are shown at the distribution layer and core layer. At the distribution layer, there are two distribution layer switches, the minimum required to support redundancy at this layer. The access layer switches, S1, S3, S4, and S6, are cross-connected to the distribution layer switches. This protects your network if one of the distribution switches fails. In case of a failure, the access layer switch adjusts its transmission path and forwards the traffic through the other distribution switch.

Some network failure scenarios can never be prevented, for example, if the power goes out in the entire city, or the entire building is demolished because of an earthquake. Redundancy does not attempt to address these types of disasters. To learn more about how a business can continue to work and recover from a disaster, visit: http://www.cisco.com/en/US/netsol/ns516/networking_solutions_package.html

Start at the Access Layer

Imagine that a new network design is required. Design requirements, such as the level of performance or redundancy necessary, are determined by the business goals of the organization. Once the design requirements are documented, the designer can begin selecting the equipment and infrastructure to implement the design.

When you start the equipment selection at the access layer, you can ensure that you accommodate all network devices needing access to the network. After you have all end devices accounted for, you have a better idea of how many access layer switches you need. The number of access layer switches, and the estimated traffic that each generates, helps you to determine how many distribution layer switches are required to achieve the performance and redundancy needed for the network. After you have determined the number of distribution layer switches, you can identify how many core switches are required to maintain the performance of the network.