Awesome distributed transactions

A curated selection of distributed transactions protocols

Highly Available Transactions

Transactional Causal Consistency

Cure: Strong semantics meets high availability and low latency

Stronger Semantics for Low-Latency Geo-Replicated Storage (aka Eiger)

Don’t Settle for Eventual: Scalable Causal Consistency for Wide-Area Storage with COPS

Stronger than read committed (MAV & RA)

n MAV provides "cut isolation" and atomicity. RA is MAV which prevents fractured reads.

Highly Available Transactions: Virtues and Limitations

Scalable Atomic Visibility with RAMP Transactions

Weaker than snapshot isolation levels (PSI, PC-PSI, NMSI)

Compared to snapshot isolation (SI) PSI, PC-PSI & NMSI allows "long fork" anomaly.

Database Replication Using Generalized Snapshot Isolation

Transactional storage for geo-replicated systems (aka Walter, PSI)

I Can’t Believe It’s Not Causal! Scalable Causal Consistency with No Slowdown Cascades (aka Occult, PC-PSI)

Non-Monotonic Snapshot Isolation (aka Jessy, NMSI)

Blotter: Low Latency Transactions for Geo-Replicated Storage (NMSI)

Snapshot isolation

Large-scale Incremental Processing Using Distributed Transactions and Notifications (aka Percolator)

Serializability

Notes on Data Base Operating Systems (aka 2PC, two-phase commit, it all started there)

Let's put 2PC coordinator on Paxos:

• Reducing the Cost for Non-Blocking in Atomic Commitment (aka MD3PC)
• Increasing the Resilience of Distributed and Replicated Database Systems (aka E3PC)
• Consensus on Transaction Commit (aka paxos commit)

CockroachDB: The Resilient Geo-Distributed SQL Database (Parallel Commits over Paxos)

Spanner: Google’s Globally-Distributed Database (2PC over Paxos)

Concurrency control schemes:

• Extracting More Concurrency from Distributed Transactions (aka RoCoCo)

• An Evaluation of Distributed Concurrency Control

• Strong consistency is not hard to get: Two-Phase Locking and Two-Phase Commit on Thousands of Cores

• Independent transactions (aka one-shot transactions):

  • Granola: Low-Overhead Distributed Transaction Coordination
  • Consolidating Concurrency Control and Consensus for Commits under Conflicts (aka Janus)

• The End of an Architectural Era: (It’s Time for a Complete Rewrite) (serial execution, aka H-Store)

Deterministic ordering:

• Calvin: Fast Distributed Transactions for Partitioned Database Systems (deterministic transaction, OLLP)
• SLOG: Serializable, Low-latency, Geo-replicated Transactions
• Ocean Vista: Gossip-Based Visibility Control for Speedy Geo-Distributed Transactions

Blending of replication with transaction processing

• Building Consistent Transactions with Inconsistent Replication (aka TAPIR)
• Meerkat: Multicore-Scalable Replicated Transactions Following the Zero-Coordination Principle

Special hardware

The End of a Myth: Distributed Transactions Can Scale (aka NAM-DB, SI)

No compromises: distributed transactions with consistency, availability, and performance (aka FaRM, serializability)

FaSST: Fast, Scalable and Simple Distributed Transactions with Two-Sided (RDMA) Datagram RPCs

Eris: Coordination-Free Consistent Transactions Using In-Network Concurrency Control (serializability, independent transactions, deterministic transaction, OLLP)

Limits of distributed transactions

The SNOW Theorem and Latency-Optimal Read-Only Transactions

Causal Consistency and Latency Optimality: Friend or Foe?

Distributed Transactional Systems Cannot Be Fast

Distributed transactional reads: the strong, the quick, the fresh and the impossible

Testing

Elle: Inferring Isolation Anomalies from Experimental Observations