Unifying Timestamp with Transaction Ordering for MVCC with Decentralized Scalar Timestamp
summary
DST is a decentralized scalar timestamp scheme for scaling distributed transactions under MVCC that avoids a centralized timestamp service. It piggybacks timestamp ordering on existing concurrency-control protocols (e.g., OCC, 2PL) by deriving a serializable order from conflicting tuples, and uses a hybrid scalar timestamp (physical clock + logical counter) to provide snapshot reads with bounded staleness. The paper gives integration guidelines and shows DST improving throughput and latency in three systems (DrTM+R, MySQL cluster, ROCOCO) across TPC-C and TPC-E benchmarks.
AI-written summary · about
chapters (5)
- 0:00Timestamp and MVCC introduction
- 3:23Decentralized timestamp challenges
- 6:33Scalar timestamps solution
- 9:59Implementation and results
- 11:07Summary and outlook
related talks
16mNCC: Natural Concurrency Control for Strictly Serializable Datastores by Avoiding the...
41mTransactions: myths, surprises and opportunities
45mTransactions and Concurrency Control Patterns
51mHigh-concurrency distributed snapshots
14mVerifying vMVCC, a high-performance transaction library using multi-version concurrency..
45mImportance of Transactional Metadata (Or How We’re Improving Upon Gossip)
49mThe Paradigm Shift - Understanding the Move To Distributed Databases
24mDistributed Transactions at Scale in Amazon DynamoDB