The Cost of Storing Checkpoints to Multiple Volatile Storage Locations Using at-least-k Semantics

Using the volatile memory of other computers within a system to store checkpoints is an alternative to the traditional approach of using stable storage. The objective of this study is to develop a storage mechanism using at-least-k delivery semantics. This semantics allows data to be saved to a minimum number of computers simultaneously using group communications, without requiring that each group computer successfully acknowledge the receipt. The new storage mechanism is implemented in the GENESIS checkpointing facility v2.0. The results showed that at-least-k storage mechanisms provide low storage latency times; however, the incurred execution overheads on the applications executing within the system are higher than that when using remote stable storage to store checkpoints.