From: Distributed systems

One approach to providing fault tolerance in distributed systems
involves the use of redundant services, such that standby facilities
can become active in the event of the failure of, or loss of
connection to, a primary service.

Another approach is to provide multiple paths of connectivity between
the computers that make up the distributed system.  The QNX system,
for example, supports multiple network drivers per node.  The purpose
of the network connection under QNX is to merge the microkernels on
the LAN into a single logical kernel.  Hence, if multiple LAN
connections per node are present, the networking code can load balance
the LAN traffic on the paths available.  It can also route around
failed links, providing both greater LAN bandwidth and better fault
tolerance.

See below for treatment of fault tolerance in systems which make use
of distributed shared memory.