The recovery problem in distributed systems

It is argued that the conventional transaction concept is viable in a general distributed system only with respect to its all-or-nothing property. In developing a recovery strategy, methods for damage assessment also have to be considered. This means more overhead, especially in the recovery phase, but the penalties for normal, i.e. error-free processing, can be alleviated. The ultimate goal should be to provide a set of tools comprising and combining mechanisms for synchronization, recovery, and even dynamic reconfiguration after failures from both strategies. This should allow the user flexible use of this tool kit, depending on the specific application which he is running on the system. A brief description of how some of the proposed concepts have already been implemented in the first prototype of the PROFEMO distributed system is included