RSS
  Multihoming Your Internet Service With BGP The administrative weight is usually assigned manually by the network administrator and can be applied to effect a degree of preference. It can be based on any number of criteria and is local to the router. This is an important point to consider, since administrative weights are not sent out in BGP routing updates. A weight can be a number from 0 to 65535. Paths that the router originates have weights of 32768 by default, other paths have a default weight of zero. If you have certain neighbors that you want to prefer for most of your traffic, you can assign a weight to all paths learned from the preferred neighbor. If you are going to manually administer weights or preferences, it is best to apply routes received from a particular neighbor, rather than individual routes since maintaining these values can become onerous. Local preference defaults to a value of 100, and unlike administrative weight, is sent in route updates to other Border Gateway Protocol routers within the same Autonomous System. Local preference is also set by the network administrator in the router configuration. An IGP route update message, such as that sent by RIP or IGRP, is a broadcast using User Datagram Protocol (UDP) as the transport protocol. Since BGP communication occurs only between peer routers, TCP is used as the transport protocol to add a measure of reliability to route update delivery. When configuring a Border Gateway router, it is necessary to define an IP address for every peer that is to receive route updates. It is recommended that you configure a loopback address on the destination router for this purpose. A loopback address is up for as long as the router is up; if you select an interface address on the destination router, it will not be able to receive updates i f this interface is down. What BGP Buys You There are basically two options to consider when implementing multiple Internet links (see "Example of BGP Configuration," on page 121). One is to use BGP to exchange route information among entities; the second is to use a static routing configuration in the entities' routers. Given that our goal is to split Internet traffic from the same router across two separate links going to two different ISPs, we can also consider these options as two distinct problems--one for sharing outbound traffic and one for sharing traffic inbound from the Internet. Outbound traffic is simpler to control, so let's look at that first. The easiest way to split outbound traffic between two links is not to use BGP at all, but to define two equal cost default routes for the router connecting your site to the Internet. In the "Example of BGP Configuration," diagram on this page, one default route will point to ISP A and the other will point to ISP B. The router will then balan ce outbound traffic to the Internet across the two links regardless of ultimate destination. The advantages to this type of solution are as follows: simple configuration, equal us e of outbound bandwidth across the two links, automatic failover in case one link goes down, and the router does not need an expensive processor, software or lots of memory. The downside is if the ultimate destination on the Internet is closer to one ISP than the other, you have no control over which link the outbound traffic will travel over. TCP will have to handle out-of-sequence packet delivery, and trouble on one ISP network will affect all your traffic. The second option that uses BGP to share route information among entities requires a lot of router memory, since it has to hold tens of thousands of entries in its routing table. Are You Full? When implementing BGP, there are a few choices that can be made with respect to the number and nature of the routing updates sent to the organization by the ISPs. If you de cide to use BGP, your ISP will request that an extensive questionnaire be completed. You can choose to have the ISP advertise only specific routes via BGP to your router, or to all Internet routes known to the ISP. The option to have all known Internet routes sent to you is termed full-BGP routing. With full-BGP routing, both ISPs will send the entire Internet routing table. Your router must decide which one of the two possible entries for each Internet destination (one from each ISP) it will put in its routing table. The router will choose the route offering the least cost to a remote network. This allows a fine degree of control over which link is used for outbound traffic to destination networks. |
 |
 |
Finding the Holy Grail o f Modem Connectivity by Chris Lewis Updated July 10, 1997 |
