The availability of a TCP/IP network design is measured by the percentage of time users are able to transfer data by using the TCP/IP network
infrastructure.
To improve the availability of a routed network, your design must include redundant links and routers between locations where the highest
availability is required. The inclusion of redundant routed paths in your design improves:
- Availability
- Performance (bandwidth)
Multiple gateways are important on segments that must always have an available inbound and outbound route.
For example, you may have configured a DMZ for your Web servers that you use for e-commerce or partner access.
A single gateway has no fault tolerance because if that single access point fails, no inbound or outbound communication can move into, or out of, the DMZ segment.
You can configure cost metrics for the routed paths to constrain traffic to a single link where the backup is a dial-on-demand interface, or allow concurrent use of multiple routes where they are available.
For example, you may have configured a VPN link on your T1 connection that allows you to connect to other sites within your company. However, for fault tolerance, you have decided to provide a demand-dial ISDN route to
critical sites. You would set the cost metric higher on the ISDN route so that the T1 route is used preferentially.
Set the link cost metric for each route to be equal if the links are available concurrently and there is no monetary cost penalty. Set different
cost metrics if a link provides dial-on-demand backup and monetary costs prohibit running multiple links concurrently.
Routers can provide built-in load balancing across multiple links and paths if the routes are concurrently available and advanced routing protocols are used.
Load balancing is possible with unequal cost metrics if protocols such as
Interior Gateway Routing Protocol (IGRP)[1] are used.
The router uses the cost metrics to determine which paths the packets take, and the administrator can adjust the amount of load balancing. This
is considered only if the speed of the redundant links differs greatly, and the links are concurrently available.
The next lesson wraps-up this module.