Open sirinath opened 10 years ago
Yes, you should use OffHeap Memory support. See GridCacheMemoryMode.OFFHEAP_TIERED or OFFHEAP_VALUES.
What my concern is does GG use this internally. Have you go benchmarks on how long a GG cluster can run without a GP pause in any of the nodes.
If you allocate little amount of heap (e.g. 2GB), and large amount of offheap (e.g. 200GB), GC is not a factor. We have internal benchmarks to confirm this.
Say I have high utilisation (60% to 70% of heap). Best is that you move to object pooling / off heap internally to make sure you you don't hit pauses and you can always schedule to run System.gc()
at a off peak time. Ideally it should be that you can afford to create GCable object on startup and shutdown only and non when you are live.
These are un related but perhaps you can look at: https://github.com/openhft https://github.com/reactor/ http://javolution.org/
You need hard realtime commitments.
Is it possible to make sure GG does not get GC pauses or gitter during operations.