Using this pattern, we allow each host to only serve a particular type of microservice. The main benefit is that you can tweak the machine to better fit the desired workload and services are well isolated. When you provide extra-large memory or fast storage, you'll be sure that it is used only for the microservice that needs it. The service is also unable to consume more resources than provisioned.
The downside of this approach is that some of the hosts may be under-utilized. One possible workaround is to use the smallest possible machines that still satisfy the microservice requirements and scale them when necessary. This workaround, however, does not solve the issue of additional overhead on the host itself.
