Abstract:

The paradigm of near-data computing has driven banks and securities firms to build multiple data centers globally or nationally.In the traditional business model, transactions focused on accessing data within a single data center.With the changing business model, distributed transactions across data centers have become common, such as transferring money between bank accounts or exchanging equipment between game accounts, with data stored in different data centers in different regions.Distributed transaction processing requires the two-phase commit protocol to ensure the atomicity of the sub-transactions submitted by each participating node.In processing cross-domain transactions, traditional transaction processing technology needs to be expanded to ensure that the system can provide higher throughput due to the longer and more varied network latency between nodes.After analyzing the problems and optimizing space for crossdomain distributed transactions, this paper proposes a new distributed transaction processing algorithm called Harp.Harp delays the execution of some sub-transactions based on the difference in network latency while ensuring serializable isolation level, reducing the duration of transaction lock contention, and improving system concurrency and throughput.Experiments show that Harp improves the performance by 1.39 times compared with the traditional algorithm under YCSB workload.

Key words: cross-domain distributed transaction, network difference, transaction scheduling, lock contention