We did a similar migration last week, but with a mail server.
You can't get transparent failover without clustering etc. - existing connections will be terminated when the system goes offline. So it's really a case of reducing the period of unavailability to the minimum.
What we did was give the replacement box a different IP address for the commissioning process but give it the same IP and hostname when it went live. Since changes don't take effect until you restart networking you can change the settings of the replacement to the live configuration, shutdown the existing box and immediately restart the networking on the replacement to bring it into service.
So the downtime became the time taken to replicate any data from the old box to the new and reset permissions etc. (I wrote a small shell script to automate this bit), plus a minute to get the new box to the live network settings.
Any changes to DNS make the exact duration of service unavailability a bit uncertain because of caching and time delays on replication. By keeping the DNS record exactly the same you avoid the issue.
Last edited by hob; 09-11-2004 at 06:44 PM.
|