11/26/14 1
Fault Treatment
•In this phase, the fault is first isolated and then repaired. The repair procedure
depends on the type of fault. Permanent faults require that the failed
component be replaced by a non-failed component. This requires a standby
component. The standby component has to be integrated into the system,
which means that its state has to be synchronized with the state of the rest of
the system. There are three general types of standby schemes:
•Cold standby -- This means that the standby component is not operational, so
that its state needs to be changed fully when the cutover occurs. This may be a
very expensive and lengthy operation. For instance, a large database may have
to be fully reconstructed (e.g., using a log of transactions) on a standby disc.
The advantage of cold standby schemes is that they do not introduce overhead
during the normal operation of the system. However, the cost is paid in fault
recovery time.
•Warm standby -- In this case, the standby component is used to keep the last
checkpoint of the operational component that it is backing up. When the
principal component fails, the backward error recovery can be relatively short.
The cost of warm standby schemes is the cost of backward recovery discussed
earlier (mainly high overhead).
•Hot standby -- In this approach, the standby component is fully active and
duplicating the function of the primary component. Thus, if an error occurs,
recovery can be practically instantaneous. The problem with this scheme is that
it is difficult to keep two components in lock step. In contrast to warm standby
schemes, in which synchronization is only performed during checkpoints, in this
case it has to be done on a constant basis. Invariably, this requires
communications between the primary and the standby, so that the overhead of
these schemes is often higher than the overhead for warm standby.
