In message <35235842.537A2574at_private> you write: > Company A now wants to improve resilliance. The have datacenters in > three > continents and so the basic idea is to put up three copies. Now the dns > entry will > point to one of them, if that fails then the contents of the dns will be > changed (not > by hand) to point at the secondary etc. Use a very short ttl on the dns > entry and > things should start again after a short while. What we've done in a similar project for an intranet at a large company: * give all servers the same (virtual) IP address (192.168.0.1) * announce the availability through routing protocols There's one A record for the service (no need to mess with these) and all clients will use the ``closest'' box (as defined by the routing metrics). If one box fails, the routes will no longer be propagated through the net and clients will be redirected to the other servers (time depends on the routing protocols used). This will happen more or less transparently ... persistent connections will fail, but (eg.) HTTP accesses should be fine. Of course, you'll need other mechanisms to make sure your services are synchronized. Hope this helps, \Bernhard.
This archive was generated by hypermail 2b30 : Fri Apr 13 2001 - 12:54:04 PDT