Hi,

We will have customers in Europe, Latin America and US, we are looking for mirroring our server in Europe, we want to find a solution to split the traffic between the server in US and Europe (the round robin trick is not good for what we have in mind), is it possible to control which server will deliver a request from a user by monitoring the response time, example:

A web surfer in France want to go to monpays.com, we will have this site on serverUS and serverEurope, for the high cost of bandwidth in Europe we prefer to deliver from serverUS but sometime the transatlantic connection is overload or it is the peak time in US and the response time go higher, and then we will deliver from serverEurope if the response time is superior to 150 ms for example, any way to setup that? Any suggestion?

Thank you for your help ;)

Jacques

I suggest you do a research on High availability solutions and what your options are.

Right off the bat, it is going to cost you money . . .

You can only save $ on bandwith if you host in US. Bandwith is cheaper in US only.
How will you check the response time if your internet connection is bad?

What I want to setup, is when the response time or the overload of the backome decrease the accesibility to the server located in US, the server located in Europe start to delivery to the surfer in Europe.

I hope that the serverEurope will work only for 20% of all the request from Europe.

You don't happen to be broadcasting BGP routes, do you? I imagine not... which is unfortunate, since this would be the easiest way of doing this.

The problem essentially comes down to deciding, given an IP address, if that computer is in Europe. It is easy to switch between two different DNS data files (one which points everyone at your north american server and one which points some people at north america and others at Europe), but you need to set up that second file so that people get the correct response based on their locale.

If you were broadcasting BGP routes, I'd suggest broadcasting routes for three IP ranges, call them block0, block1, block2. Server1 should broadcast BGP routes for block0 and block1; server2 should broadcast BGP routes for block0 and block2. This would essentially turn block0 into an anycast block, with packets going to whichever server is closest. Naturally, since stateless anycasting requires the use of stateless protocols, you could only use block0 for DNS, thus the use of blocks 1 and 2 for stateful protocols (eg HTTP). But since you're talking about only 2 servers you probably don't run BGP, so this paragraph can be completely disregarded.