Well fault tolerance as I know
it is reliability i.e the server  is immune to
failures and can react to failures and correct itself
automatically.<br><br>this is done using the familiar aphorism "redundancy
increases reliability". the server actions are duplicated
and exist on say multiple computers as components.
say the same task T is implemented by two different
components A and B. now the server always uses component A
to service the task T. Now if A fails then the
server can use B for implementing T.<br><br>This is
usually done using RMI/CORBA which support distributed
objects.

In Java and particularly RMI, I interpret tolerance
to mean that an error can be detected and your code
deals with it in some that allows the code to keep
running (maybe retry or offer a dialog to the user to
submit different values, etc).<br><br>This means that
you code try/catch blocks all over your code and you
handle all of the possible exceptions and errors that
can be thrown. You don't ever let the program just
quit without the user instructing it to do
so.<br><br>RMI requires that you catch RMIExceptions. What you
do with these is what determines if your code is
fault tolerant, or not.