3 ms·
No, the test is more than that. The preallocated objects matter because they are pathologic cases for certain GC operations. Large allocations compound this fur
by the8472 7y ago
No, the test is more than that. The preallocated objects matter because they are pathologic cases for certain GC operations. Large allocations compound this further.
I guess CMS manages to perform fine here because it can flexibly resize the young generation and isn't region-based while G1 has special optimizations for what they call humongous allocations.
It's a test that checks how collectors perform under two combined deviations from the assumptions for which they're optimized (i.e. that object graphs have a high branching level that allows parallelization, short-lived objects are small on average)