EBGP load balancing

[netuser786]

hi,
I am a beginner in networking area, while reading about the BGP rules i came to know about the longest prefix match.in this regard, i thought if we have:


138.39/16 / ISP3 \ 138.39.1/24
/ \
/ \
ISP1 _____P2P_____ISP2 138.39.1/24
138.39/16 \ /
\ /
CUSTOMER
138.39.1/24


i saw these slides online while doing search for BGP. according to this figure ISP3 will send the traffic to customer thru ISP2 due to the longest prefix match rule.ofcourse, ISP1 and ISP2 both are providers of the customer. i want to ask that if the network route thru ISP2 becomes congested or have higher load, but since due to the longest prefix match ISP3 will still send data through ISP2, how can the load be balanced in this situation?
i found a solution to this which says split the prefix in two, to balance the load. would that work?
can there be some other way to balance the load while keeping the BGP rules valid??:\:\

 

[JackMDS]

http://www.cisco.com/en/US/tech/tk365/technologies_configuration_example09186a00800945bf.shtml

 

[drebo]

Short answer is that you can't.

Long answer is that BGP engineers are paid great sums of money to design these networks.

Basically, under the most basic configuration, BGP behaves exactly like RIP--it uses a strict hop count to determine a path. The hop count is not the number of routers, but rather the number of autonomous systems between you and the destination prefix. It is not possible to know what is going on in the downstream autonomous systems, however, so there's little that you can do to circumvent congestion in someone else's network.

There are a variety of metrics that the customer could use to influence upstream routing, from communities (depending on providers) and AS Path padding. For your part, you could configure your router to multipath. This may or may not be a good idea, depending on what it is you are actually trying to accomplish, as it can have some other interesting effects.

Basically, unless there is a very, very specific goal you need accomplished, it's unlikely that you're going to properly be able to loadbalance your connections TO someone else's network through any number of third-party networks. At best, the destination could influence the route the rest of the world takes to get to them.

 

[Gryz]

As a rule of thumb, all routing protcols in the TCP/IP protocol suite do never take load or congestion into consideration when computing routes.

There is technology to adjust your traffic flows according to load and congestion. Those technologies are called "traffic engineering". Strictly speaking, this is not done by the routing protocols, but by other parts of software. E.g. there is something called "MPLS Traffic Engineering" that can dynamically alter the flow of traffic through your network. But this is pretty advanced technology, and I suggest a beginner would not bother learning about it yet. There are many other subjects to learn which are just as interesting.

 

[Lithium381]

Learn about the IGP's first(RIP, OSPF, maybe even EIGRP) and then get your feet wet with BGP. It's a very "simple" protocol but with LOTS of power for making it behave how you want.

In your case, you seem to be looking for a routing protocol that will dynamically switch routes when one route gets too congested. The only one that could do that off the top of my head would be EIGRP which is a cisco proprietary protocol and not suited for inter-AS routing on a large scale.

You can definitely tweak the routes and "load balance" and force some destinations to be reachable through a sub-optimal route to alleviate some of the congestoin on a particular link using BGP though.

Also, try using the <code> tags to make your diagram more readable . .