H-store: a high-performance, distributed main memory transaction processing system
Proceedings of the VLDB Endowment
Information Sciences: an International Journal
An evaluation of checkpoint recovery for massively multiplayer online games
Proceedings of the VLDB Endowment
Hybrid in-memory and on-disk tables for speeding-up table accesses
DEXA'10 Proceedings of the 21st international conference on Database and expert systems applications: Part I
A latency and fault-tolerance optimizer for online parallel query plans
Proceedings of the 2011 ACM SIGMOD International Conference on Management of data
RAFT at work: speeding-up mapreduce applications under task and node failures
Proceedings of the 2011 ACM SIGMOD International Conference on Management of data
| Hi-index | 0.00 |
Checkpoint of an in-memory database is the main source of a persistent database image surviving a software crash, or a power outage, and is, together with transactions logs, a foundation for transaction durability. Since checkpoints are created simultaneously with transaction processing, they tend to decrease database throughput and increase its memory footprint. Of the current methods, most efficient are the fuzzy checkpoint algorithms that write dirty pages to disk and require transaction logs for reconstructing a consistent state. Known consistency-preserving methods suffer from excessive memory usage or a transaction-blocking behavior. In this paper, we present a consistency-preserving and memory-efficient checkpoint method. It is based on tuple shadowing as opposed to known page shadowing methods, and rearranging of tuples between pages for minimal memory usage overhead. The method's algorithms are introduced and both analytical and experimental analysis of the proposed algorithms show significant reduction in the memory usage overhead, and up to 30% higher transaction throughput compared with a fuzzy checkpoint method with undo/redo log.