Can be more efficient, but it’s not always. You can scale those services more too. Do you need all 16 running on idle? Lazy load then as needed. Also does it really need a whole 500mb? Why? If its just a cache made that’s better handles with another shared service (redis, etc). If it’s software, why? A full “fat” Suse Base Container image uncompressed is 94.8mb and not all of that will be loaded in RAM. Going down to a micro at 23mb. All this and now you can deploy it on the shared infra, no separate OS, Management/security/logging/networking solution.
Can be more efficient, but it’s not always. You can scale those services more too. Do you need all 16 running on idle? Lazy load then as needed. Also does it really need a whole 500mb? Why? If its just a cache made that’s better handles with another shared service (redis, etc). If it’s software, why? A full “fat” Suse Base Container image uncompressed is 94.8mb and not all of that will be loaded in RAM. Going down to a micro at 23mb. All this and now you can deploy it on the shared infra, no separate OS, Management/security/logging/networking solution.
Sure, but they also have duplication, IE database connections. Each service needs at least 1 connection versus one for a monolith.
There are pros and cons to any architecture.
The IPC increase is the most sure tradeoff for sure.
Agreed, engineering is deciding what trade offs are acceptable. Cargo culting any pattern is a good way to build runaway junk