On 5 February 2010 18:05, evebill8 <evebi...@hotmail.com> wrote: > Cool! I just want to confirm if the rule is right. My IT guy also does > not > believe it. > > It wasn't a bad rule of thumb as a place from which to start tuning when typical server memory sizes were 0.5G to 2G - it reserved "sort-of enough" RAM for the OS and the non-heap parts of Java and "sort-of enough" RAM for the web application, though as you can imagine that varied quite a lot. But it could never really be said to be optimal, and ideally you'd always measure and tune.
These days, there's so much more RAM in most servers that tuning becomes more relevant if the load's high. If your application is very heavy on disk reads, for example, you might do better to reduce its heap size in order to give the OS more space to cache disk pages. By contrast, I have one *very* heavy compute application (not directly a web app) that likes all the RAM it can get as heap memory to cache intermediate results in case they can be re-used. - Peter