The possibility to distribute every component on multiple servers will ensure scalability for very large models.
One Master could distribute read requests to multiple workers and aggregate the results from all workers before sending the result to the client.
One Master could distiribute write requests to multiple workers to enhance loading performance.
