Most enterprise database environments have a hub and spoke or snowflake architecture. There is a central transactional database server, and then there are data marts and other line of business databases that feed from or integrate with it. It's better when the spokes can still be fully or mostly operational even while the central hub experiences unplanned down time. It depends, but if some predictable latency is tolerable, then periodically replicated or merged local tables can insure better overall availability when compared to and architecture that is tightly integrated with linked server connections. If you're leveraging micro services, then they should feed from data marts or replicas.