My impression has always been that the sizing guidelines are somewhat spitballed and mostly the result of a reverse Scottyfication. That's Scotty from Star Trek.
"I think people should have 24 gigabytes to be safe for all scenarios"
"Would 32G make that even more safe?"
"Okay, then it's settled. Write down 48G."
To some extend that's understandable since everyone's requirements are different and their recommended setup is just a guideline. It's also my experience that Automic has always tried to avoid making any truly binding statements when it comes to a recommended server distribution. Consultants have to have an opinion, they're paid and it would be bad if they didn't.
So in summary I do think people should feel free to deviate from Automics guidelines when they understand what that means. You already outlined the pros and cons already, it mostly comes down to
pro:
- fewer servers to maintain, patch, power
- better utilization of existing ressources (RAM, $$$).
con:
- one runaway application affects the other one (though the GUI can often times affect the entire installation badly anyway, e.g. by blocking out all the CP even with two servers)
- downtimes of one component affect the other
- (possibly) some day some Automic rep might use it as a reason not to follow through on an issue of support
But I see no inherent problems with bunching it up on one server. In fact, while we separate engine and AWI for "prod", we have put both components on the same server for "test" and "dev" also.
Br,