Map-Matching service on world map. scalability and microservices

Hi there,
for what I saw looking the code I assume that the map-matching is performed on a machine where the full graph is stored in memory. But if we would like to host a map-matching service working worldwide doeas it mean that we have to create microservices hosting each of them the full world graph? Willing to scale horizzontaly based on the load of the requests, how long will it take to deploy a new microservice having into its belt the full world graph? Is the graph hosted into the RAM of the worker computational node? Or is it stored into a distributed RAM layer a la REDIS?
Somehow this is the same issue I would ask also for routing worldwide. Essentially I’m interested in understanding how the graph map is handled, if all in RAM, or tiled and stored in different locations or else …