We found that with the 272MB config, we were always running low on memory, and swapping in the console even without any agents or additional software. Add some agents or monitoring software and this becomes a real problem. Our default? 800MB console. Always.
Before starting with ESX 3.0.1, we contacted VMware and had some lengthy discussions about using RAM configuration. We went over in painful detail exactly what we do, and they recommended that unlike 2.5.x, an 800MB console would not be needed, 272 was the perfect number. We continued to swap, and crashed ESX a few times. Since moving to 800, ESX has returned to a stable condition.
We did run 512MB in some of our smaller systems with success for a few months, that worked well enough. I did not continue to persue this option because I wanted to run audits across hundreds of ESX servers, and it was easier to just check for 800, rather than build in exceptions for my audits. call me lazy, but 800MB works quite well and while swapping can occur, it is rare.
