Fix the accounting for gen 1 allocations by taking into account what … #60248
Conversation
…has been already consumed in this GC. Details: In gen 1 GCs, we recompute the budget at the end of the GC. However, the amount consumed in the current GC is nowhere taken into account. That causes us to do at most alternate gen 0 and gen 1 GCs, we cannot do back-to-back gen 1 GCs. If we take the gen 1 budget as the amount of memory we are aiming to be used for gen 1, then it seems illogical to just drop the amount used in this GC on the floor. The fix is to actually subtract the amount of gen 1 budget used in this GC from dd_new_allocation for gen 1. This is a bit awkward for server GC, because we even out the budget and the amount to subtract from dd_new_allocation across heaps after having computed it for each heap individually.
|
Tagging subscribers to this area: @dotnet/gc Issue Details…has been already consumed in this GC. Details: In gen 1 GCs, we recompute the budget at the end of the GC. However, the amount consumed in the current GC is nowhere taken into account. That causes us to do at most alternate gen 0 and gen 1 GCs, we cannot do back-to-back gen 1 GCs. If we take the gen 1 budget as the amount of memory we are aiming to be used for gen 1, then it seems illogical to just drop the amount used in this GC on the floor. The fix is to actually subtract the amount of gen 1 budget used in this GC from dd_new_allocation for gen 1. This is a bit awkward for server GC, because we even out the budget and the amount to subtract from dd_new_allocation across heaps after having computed it for each heap individually.
|
ghost
locked as resolved and limited conversation to collaborators
2 participants
…has been already consumed in this GC.
Details:
In gen 1 GCs, we recompute the budget at the end of the GC. However, the amount consumed in the current GC is nowhere taken into account. That causes us to do at most alternate gen 0 and gen 1 GCs, we cannot do back-to-back gen 1 GCs.
If we take the gen 1 budget as the amount of memory we are aiming to be used for gen 1, then it seems illogical to just drop the amount used in this GC on the floor.
The fix is to actually subtract the amount of gen 1 budget used in this GC from dd_new_allocation for gen 1. This is a bit awkward for server GC, because we even out the budget and the amount to subtract from dd_new_allocation across heaps after having computed it for each heap individually.