The Consistent File-Status in a User-Triggered Checkpointing Approach
Geert Deconinck, Johan Vounckx, Rudy Lauwereins · 1995
The user-triggered checkpointing tool implements a non-blocking, co-ordinated (global) checkpointing method, where the programmer defines the contents and the position of the recovery-line. Within this tool, we developed and implemented file-checkpointing. This allows to include the status of files into the checkpoints and to restore this status when the application is rolled back. The file-status in the checkpoint-data is limited to those attributes of the files that are strictly needed to guarantee consistency. The developed approach is suited for files that are accessed by a single process or shared by several processes and are accessed for reading and/or for writing. Consistency is guaranteed under the fail-silent as well as under the fail-time-bounded fault-models. The programmer-involvement is compensated by the smaller time-overhead. Keywords file-checkpointing, fault-tolerance, checkpointing and rollback, backward error recovery 1. Introduction Several families of techniques...