Removing single points of failure.
WHAT TO IDENTIFY
Every component whose failure affects customers.
WHAT COMMONLY IS A SINGLE POINT
The nameservers, if not diverse The mail server The control panel's licensing or authentication Shared storage The upstream connection The database, for a shared platform
WHAT TO DO ABOUT NAMESERVERS
Several, on separate networks and in separate locations.
WHAT TO DO ABOUT MAIL
Secondary mail servers accepting and holding mail when the primary is unavailable.
WHY THAT MATTERS
Mail is otherwise deferred by senders, and eventually returned.
WHAT TO DO ABOUT WEB
Redundancy at the application level, or rapid restoration from backup.
WHAT TO BE HONEST ABOUT
Shared hosting on one machine is not highly available, and marketing should not imply otherwise.
WHAT TO OFFER CUSTOMERS WHO NEED MORE
Arrangements that genuinely provide it, priced accordingly.
WHAT TO TEST
Failure of each component, deliberately.
WHAT TO MEASURE
How long restoration actually takes.
WHAT TO PUBLISH
What you actually commit to, and what the remedy is.
