I wonder if I'm the only one who has problems with a server which is unavailable for some time. I experienced unpredictable answering times for my calls to the server and did some research on it. I found the garbage collector to be the main reason for unpredictable response times. For servers with memory above 1GB Java VM needs rest for more than one second and more, with the low pause collector. I read from servers unavailable for minutes. For user interactions one second is hardly acceptable, minutes is unacceptable.
Think about a database that needs 30 seconds to rearange its cache, would you accept that? Would you accept an identity system with door opener which normally opens within milli seconds and sometimes you smash the door because it needs to collect garbage.
Oh, I hear them screaming, what you want is a real-time system. Yes and no, yes I want predictable response times and no I don't need real-time threads or similar.
I want the behaviour of the common databases, a small request is a small request and its fast as long as the server doesn't run out of resources.
The current Java situation is a step back and needs attention.

Have you ever heard of concurrent (Concurrent mark-sweep) GC ?

The CMS collector executes "mostly concurrently" with the application, hence, is sometimes referred to as the "mostly-concurrent garbage collector". It trades the utilization of processing power that would otherwise be available to the application for shorter garbage collection pause times.#

Turbo-charging Java HotSpot Virtual Machine, v1.4.x to Improve the Performance and Scalability of Application Servers

My experience was with the concurrent garbage collector and after reading the SUN documatation and long tuning sessions. It's not that easy to solve it.