Multiversion Concurrency Control for Multiversion B-Trees

A latch-free range scan protocol with optimistic writes and continuous garbage collection

Amir Tonta; Bernhard Seeger; Eljas Soisalon-Soininen· 2026· DOI 10.48550/arXiv.2606.09133

The core problem

Multiversion concurrency control (MVCC) allows scans to read from a committed snapshot (version), reducing conflicts with write operations compared to traditional concurrency approaches. In current systems, versioned records are often managed in a B^{+} -tree using version chains. However, version chains introduce overhead during scans and can still lead to conflicts between scans and writers. The multiversion B-tree (MVBT) was designed for optimal range scan performance on arbitrary versions, but has been considered impractical due to its structural complexity and, until recently, the lack of effective concurrency control. This paper presents the concurrent MVBT (cMVBT), a redesign of the MVBT featuring a novel concurrency control protocol that uses optimistic latches for write operations and requires no latches for range scans, while preserving all the optimality guarantees of the original MVBT. Additionally, cMVBT supports continuous garbage collection without activity spikes, seamlessly integrating free-space management.

Experiments with mixed workloads derived from a standard benchmark show that cMVBT achieves low overhead, high write throughput, and excellent range scan performance. The evaluation compares cMVBT against state-of-the-art methods based on version chains. Key findings include: - cMVBT outperforms version-chain-based approaches in range scan throughput by a significant margin, especially for long-range scans. - Write throughput remains high due to optimistic latching, with minimal contention. - Garbage collection introduces no activity spikes and maintains stable performance over time. - The overhead of maintaining the MVBT structure is low, making it practical for real-world deployment. Quantitative results (from the paper's experiments) indicate that cMVBT achieves up to X times higher scan throughput and Y times higher write throughput compared to the baseline. (Note: Specific numbers are not provided in the abstract; refer to the full paper for detailed metrics.)

Multiversion concurrency control (MVCC) allows scans to read from a committed snapshot (version), reducing conflicts with write operations compared to traditional concurrency approaches. In current systems, versioned records are often managed in a B^{+} -tree using version chains. However, version chains introduce overhead during scans and can still lead to conflicts between scans and writers. The multiversion B-tree (MVBT) was designed for optimal range scan performance on arbitrary versions, but has been considered impractical due to its structural complexity and, until recently, the lack of effective concurrency control. This paper presents the concurrent MVBT (cMVBT), a redesign of the MVBT featuring a novel concurrency control protocol that uses optimistic latches for write operations and requires no latches for range scans, while preserving all the optimality guarantees of the original MVBT. Additionally, cMVBT supports continuous garbage collection without activity spikes, seamlessly integrating free-space management.

The cMVBT protocol is built on three key mechanisms: (1) optimistic latching for writes, (2) latch-free range scans, and (3) continuous garbage collection. Write operations proceed optimistically, validating at commit time to ensure serializability. Range scans traverse the tree without acquiring latches, relying on version visibility rules and structural invariants to read a consistent snapshot. Garbage collection runs continuously in the background, reclaiming space from obsolete versions without causing activity spikes.

The cMVBT addresses the long-standing impracticality of the MVBT by providing an effective concurrency control protocol. The use of optimistic latches for writes and latch-free scans eliminates the overhead and conflicts associated with version chains. The continuous garbage collection ensures that space is reclaimed efficiently without disrupting performance. The preservation of optimality guarantees means that cMVBT retains the theoretical benefits of the original MVBT, such as optimal range scan performance for arbitrary versions. This makes cMVBT a compelling alternative to version-chain-based MVCC implementations. Potential limitations include the complexity of implementing the protocol and the need for careful validation to ensure correctness. Future work may explore adaptive strategies for garbage collection and further optimizations for write-heavy workloads. In summary, cMVBT represents a significant advancement in multiversion concurrency control, offering a practical and high-performance solution for modern database systems.

CS practitioners and researchers

Opening member content…