MATSim Code

Made some adjustments to try and get it running with Frans Brooks.

fill the SoftCache with random stuff to monitor GC-related trace messages

it seems the JVM is able to clean only part of the softreferences when
it needs space. See comment in script for details

better log and documentation

trace info on cache cleanup

bugfix

somehow misunderstood the return value of putIfAbsent...

let the cleanup be done by a daemon thread

rename classes: remove "lru" from names

drop the LRU mechanism and use a ConcurrentHashMap for softReferences

using the cache from several threads is currently very slow, because of
synchronization for each put(). ConcurrentHashMap has neat mechanisms to
reduce thread communication to the strict minimum.
The LRu was actually not very useful, given that soft references allow to
keep data almost indefinitely if enough RAM is given.