experiments with jdk9

[matecs]

hi,
compact strings does not add too much in my case because i don't have too
much strings, but a lot of lists of arbitrary own binary types, mainly this:
http://sources.nop.hu/src/tab/tabRouteEntry.java
there are a lot of similarities as a lot of route properties match, for example
bgp as paths and bgp communities, etc... but in generally speaking, i'm
very satisfied with the memory footprint, the full internet table contains
650k routes nowadays, i've 100 peer routers to flood this info to, and i
keep track of each route per peer (which is just references to the calculated
best path, so these differs just for a very limited time frame, when convergence
happens in the wild and the old, sent route does not match newly calculated one),
and the whole big jvms use 10-13gb of ram, and the g1gc does not pause for more than
300 msec which is within the bounds of the routing protocols' requirements (ospf).

regarding the performance, my measures show about 10% gain on my routing table
operations (add, del, replace, longest-match) which is very nice achievement!
regards,
cs


> it's great to hear that freerouter works well on JDK 9. I hope that 
> improvements in JDK 9, like compact strings, present a performance 
> benefit for the big JVMs.