Title :
Parallel checkpoint/restart without message logging
Author :
Meth, Kalman Z. ; Tuel, William G.
Author_Institution :
IBM Haifa Res. Lab., Haifa, Israel
Abstract :
We describe a parallel checkpoint/restart mechanism. The checkpoint is performed among the participating parallel tasks using a new algorithm that we call “Stop and Discard”. Tasks may be checkpointed without waiting for previously sent messages to be received. Specific message logging is not required. Message data that may be in transit is saved in the checkpoint files
Keywords :
message passing; parallel algorithms; program diagnostics; system recovery; checkpoint files; in-transit message data saving; message logging; parallel checkpoint/restart mechanism; parallel tasks; stop-and-discard algorithm; task checkpointing; Concurrent computing; Delay; Hardware; Kalman filters; Laboratories; Process design; Protocols;
Conference_Titel :
Parallel Processing, 2000. Proceedings. 2000 International Workshops on
Conference_Location :
Toronto, Ont.
Print_ISBN :
0-7695-0771-9
DOI :
10.1109/ICPPW.2000.869110
