Load Balancing under Linux

[Gerald Combs]

On Thu, 25 Apr 2002, Jonathan Hutchins wrote:

> I asked a very high level ISP tech about this, and he said they use a
> protocol similar to RIP and OSPF.  They handle a lot more traffic, and use a
> Cisco router to do the balancing.  He suggested registering for RIP and
> going with that.

A "very high level ISP tech" suggested you use RIP?  I'd find another tech
if I were you.  What does "registering for RIP" mean, BTW?

While many dynamic routing protocols offer decent failover (BGP, OSPF,
EIGRP, not RIP or RIPv2), most don't do load balancing very well.  None do
decent load balancing on asymmetric links, e.g. two T1s and a fractional
T3.  

We're using two Cisco 7200 VXRs as border routers here at work.  We have
three T1s, two from one provider and one from another.  We're peering BGP
with our upstreams so that we can fail over if any of the links go down.
However, in order to perform proper inbound or outbound load balancing, we
have to resort to things like route maps that shoot the traffic out a
particular interface based on a regular expression applied to the
destination ASN.  

BGP is a really fancy and cool protocol.  It's used to route most of the
backbone Internet traffic.  If you want do dynamic routing with different
ISPs, it's your only choice - most providers won't peer OSPF or EIGRP with
you (I guarantee they'll laugh out loud at the suggestion of peering RIP).
But it often sucks at load balancing.

Better solutions are coming down the line, like MPLS (Multiprotocol Label
Switching), but they aren't here yet for most people.