leves at caesar.elte.hu
Tue Jun 19 23:50:58 UTC 2018
On Mon, 18 Jun 2018, Clément Bera wrote:
> There are still some unlinking, but it's less common so the single
linked list for chunk of size 1 is now affordable (I don't see 75% of time
spent in free chunk unlinking like before on pathological cases).
> In selective compactor, segments are compacted using unlinking but only
segments mostly empty are compacted, so low amount of free chunks of size
1. The problem was in the sweep phase and we fixed it.
Btw, why did you decide to use a tree instead of a hash table?
I had a look at the tree implementation, and it seems to be a simple
binary tree. Is that correct?
More information about the Vm-dev