Understanding Path Control
This section introduces path control performance issues and introduces the tools available to control path selection.
Assessing Path Control Network Performance
This chapter is concerned with controlling the path that traffic takes through a network. In some cases, there might be only one way for traffic to go. However, many networks include redundant paths, by having redundant devices or redundant links. In these cases, the network administrator may want to control which way certain traffic flows.
The choice of routing protocol or routing protocols used in a network is one factor in defining how paths are selected; for example, different administrative distances, metrics, and convergence times may result in different paths being selected. As described in Chapter 4, “Manipulating Routing Updates,” when multiple routing protocols are implemented, inefficient routing may result. For example, two-way multipoint redistribution requires careful planning and implementation to ensure that traffic travels the optimal way, and that there are no routing loops.
When a network includes redundancy, other considerations include the following:
Resiliency—Having redundancy does not guarantee resiliency, the ability to maintain an acceptable level of service when faults occur. For example, having redundant links between two sites does not automatically result in the backup link being used if the primary link fails. Configuration is necessary to implement failover, and to use the backup link for load sharing if that is desired. (Even if failover is configured correctly, the redundant link may not operate when needed; for example, if it uses the same physical infrastructure as the primary link.)
Availability—The time required for a routing protocol to learn about a backup path when a primary link fails is the convergence time. If the convergence time is relatively long, some applications may time out. Thus, using a fast-converging routing protocol, and tuning parameters to ensure that it does converge fast, is crucial for high-availability networks.
Adaptability—The network can also be configured to adapt to changing conditions. For example, a redundant path could be brought up and used when the primary path becomes congested, not just when it fails.
Performance—Network performance can be improved by tuning routers to load share across multiple links, making more efficient use of the bandwidth. For example, route advertisements for specific prefixes can be advertised on one link to change the balance of bandwidth use relative to other links.
Support for network and application services—More advanced path control solutions involve adjusting routing for specific services, such as security, optimization, and quality of service (QoS). For example, to optimize traffic via a Cisco Wide Area Application Services (WAAS) Central Manager, traffic must be directed to flow through the Cisco WAAS device.
Cisco WAAS is a WAN optimization and application acceleration solution that optimizes application and video delivery over a WAN, and is illustrated briefly in the “Cisco Wide Area Application Services” section, later in this chapter.
Predictability—The path control solution implemented should derive from an overall strategy, so that the results are deterministic and predictable. For example, traffic is bidirectional by nature; for every packet that goes out, a reply typically must come back. When configuring a routing protocol to deploy a path control strategy, consider both upstream and downstream traffic. For example, changing or tuning downstream advertisements toward a server farm could adversely affect upstream traffic flows from the server farm.
Asymmetric traffic—Asymmetric traffic, traffic that flows one on path in one direction and on a different path in the opposite direction, occurs in many networks that have redundant paths. Asymmetry, far from being a negative trait, is often desirable network trait, because it uses available bandwidth effectively, such as on an Internet connection on which downstream traffic may require higher bandwidth than upstream traffic. Border Gateway Protocol (BGP) includes a good set of tools to control traffic in both directions on an Internet connection. However, in most routing protocols, there are no specific tools to control traffic direction.
In a part of a network that includes devices or services such as stateful firewalls, Network Address Translation (NAT) devices, and voice traffic, which require symmetrical routing, traffic symmetry must be enforced or the services must be tuned to accommodate asymmetry. For example, asymmetry in voice networks may introduce jitter and QoS issues. In other areas of the network, though, it might be inefficient and undesirable to try to engineer artificial symmetry.
Optimal routing in terms of network utilization within specific requirements is typically a design goal. Those requirements should be considered within the context of the applications in use, the user experience, and a comprehensive set of performance parameters. These parameters include delay, bandwidth utilization, jitter, availability, and overall application performance. Even if the routing table on the routers includes the necessary prefixes, applications might still fail if the performance requirements are not met.