Low cost solution to http loadbalancer single point of failure

I have many low end atom servers for a startup project, each server have a few public ip addresses (no private network).

I'm currently in the process of making it failsafe. There now is a load balancer in front of my http servers.

This load balancer is at this point a single point of failure, when it goes down everything becomes unreachable.

I was wondering if there is a low cost solution to remove the single point of failure?

I looked into round robin DNS, but it is quite unreliable.

In addition I looked into heartbeat which looks like a nice solution when you have access to the hardware and can connect network cables.

But in my case i can't attach network cables or add hardware. I only have root access and can install software and reconfigure Linux.

Update

Thanks for the answers, i appreciate it.

But when using heartbeat or Keepalived you need 2 network cards in your server, my server only have 1. That one network card is the main one.

I can't use the main one because when it gets high traffic the heartbeat communication may fail (and trigger shutdown). I don't want that 🙂

Is there a alternative?