Bugzilla

Bug 1903758 shows a significant regression for the usage of resident memory. It seems that re-using memory more often can result in a higher number of only partly-used (non-full) runs.

In particular our current policy for re-use seems to be position-in-memory based, which might have been an ok-policy before we did randomization, but seems improvable.

We assume that allocations for objects of the same size often happen in bursts, as well as deallocations, as a consequence of object trees or other structures being built up and later being torn down all together.
In such situations it would help to keep allocations from the same burst happen as much as possible in the same runs, such that it gets more likely to free entire runs when the corresponding free burst occurs.

Bug 1903758 shows a significant regression for the usage of resident memory. It seems that re-using memory more often can result in a higher number of only partly-used (non-full) runs (for small allocations).

In particular our current policy for re-use seems to be position-in-memory based, which might have been an ok-policy before we did randomization, but seems improvable.

We assume that allocations for objects of the same size often happen in bursts, as well as deallocations, as a consequence of object trees or other structures being built up and later being torn down all together.
In such situations it would help to keep allocations from the same burst happen as much as possible in the same runs, such that it gets more likely to free entire runs when the corresponding free burst occurs.

Bug 1903758 shows a significant regression for the usage of resident memory. It seems that re-using memory more often can result in a higher number of only partly-used (non-full) runs (for small allocations).

In particular our current policy for re-use seems to be position-in-memory based, which seems improvable.

We assume that allocations for objects of the same size often happen in bursts, as well as deallocations, as a consequence of object trees or other structures being built up and later being torn down all together.
In such situations it would help to keep allocations from the same burst happen as much as possible in the same runs, such that it gets more likely to free entire runs when the corresponding free burst occurs.

Edit: See comment 4.

Bug 1903758 shows a significant regression for the usage of resident memory. It seems that re-using memory more often can result in a higher number of only partly-used (non-full) runs (for small allocations).

In particular our current policy for re-use seems to be position-in-memory based, which seems improvable.

We assume that allocations for objects of the same size often happen in bursts, as well as deallocations, as a consequence of object trees or other structures being built up and later being torn down all together.
In such situations it would help to keep allocations from the same burst happen as much as possible in the same runs, such that it gets more likely to free entire runs when the corresponding free burst occurs.