A high-level process checkpointing and migration scheme for heterogeneous distributed systems

Reliability is a key concern of designers of distributed computing systems. Checkpointing can be used as a basis for designing resilient processes and process-migration schemes, but very few systems which implement process-checkpointing are heterogeneous. 'High-level' process checkpointing schemes capture process-state at a higher level of abstraction than do low-level schemes. The resulting state does not depend on low-level or platform-specific structures, and so is meaningful at any site in a heterogeneous distributed computing network. This paper presents a high-level approach to process checkpointing which is transparent to the programmer, which operates at a fine level of granularity, and which can deal with dynamically allocated memory and multithreaded processes.