BigMemory Further Performance Analysis

This page contains further performance results for off-heap store covering less common scenarios.

50% reads: 50% writes

In this scenario the cache.put() is called 50% of the time and cache.get() is called 50% of the time. 90% of the time 70% of the keyset is selected - i.e. a very weak hot set. Other test settings are the same. This scenario is therefore a very cache unfriendly which exacerbates GC problems. Therefore the on-heap store suffers badly as seen in the charts. However the off-heap store retains it linear behaviour. Something else this test shows is that the off-heap store does not suffer from fragmentation. On this last point, we did internal testing with a worst case scenario for fragmentation where we constantly replace with randomly sized elements. The off-heap store also handles this situation very well.

Largest Full GC

Latency

Throughput