Optimizing Distributed Transaction Performance through Adaptive Replica Provision

Xinyue Shi

doi:10.1109/tkde.2026.3733100

Optimizing Distributed Transaction Performance through Adaptive Replica Provision

2026-09-11

0

PRE

AI

DOI：10.1109/tkde.2026.3733100

Original

Original request for help

Save

Abstract

En 中文

Distributed transactions require multiple rounds of cross-node communications, leading to high latency and limited throughput. Existing approaches convert distributed transactions into single-node transactions by either migrating co-accessed partitions to the same nodes or establishing a super node that hosts all partitions. However, migration-based methods may block transactions during data migration, while super nodes risk becoming performance bottlenecks under heavy workloads. In this paper, we present Lion, a transaction processing protocol that utilizes partition-based replication to reduce the occurrence of distributed transactions. Inspired by the fact that modern distributed databases horizontally partition data, with each partition having multiple replicas, Lion aims to assign each transaction a node that hosts at least one replica of every partition the transaction needs to access. To ensure such a node is available, Lion proposes an adaptive replica provision mechanism, enhanced with a Transformer-based workload prediction algorithm, to determine the appropriate replica placement. The adaptation of replica placement is conducted preemptively and asynchronously, thereby minimizing its impact on performance. Extensive experiments show that Lion achieves up to 2.1× higher throughput and 76.4% better scalability than existing approaches. We further integrate Lion into a real-world distributed database to demonstrate its practicality.

Keywords:

Transaction Processing

Replica Provision

Optimizing Distributed Transaction Performance through Adaptive Replica Provision

Abstract

Journal

Organization

Cited Papers