Architecture Independent Fault Tolerance

In Dome, we are supplying checkpoint/restart functions that are written in the source language (C++) itself, making them easily portable to multiple architectures. Unlike traditional fault-tolerance routines, which save core memory images, Dome just saves the values needed from the higher-level data structures. A preprocessor helps to modify programs so that the stack and program counter can more easily be restored without much extra work for the user.

Advantages of Our Method

• Architecture independence
• Small checkpoints
• Simple, easily understandable implementation
• Checkpoints may be used to provide both fault tolerance and process migration.

How It Works

• dome::checkpoint() checkpoints all objects in the Dome environment.
• User's code should call dome::checkpoint() on each iteration.
• On restart, dome::restart() restores dome variables.
• A preprocessor modifies programs to save the stack and program counter.

Status

• Implementation is currently in progress. See our latest tech report, linked in from the Dome home page.