A Low-Latency Network Design Scheme for Distributed Databases With Application-Specific Access Distributions

This letter proposes a network design scheme for distributed databases leveraging application-specific access distributions known before deployment. By considering access distributions and allowable data staleness, the scheme optimizes communication paths to minimize total operation latency. We formulate this as a mixed-integer linear programming (MILP) problem, optimizing leader placement, quorum configurations, and replica-holding nodes. Numerical results on the JPN25 and NSFNET topologies show that the proposed scheme reduces the average latency by up to 84.6%, with an overall average reduction of 40.8%, compared with the benchmark schemes. These results demonstrate that incorporating application-specific access distributions into network design enables low-latency data access, even from service launch, without requiring usage data.