Closed qinsoon closed 4 years ago
In GitLab by @caizixian on Feb 10, 2018, 14:01
Yi's ISMM2016 paper shows that having a central controller thread to receive works and push them to the global deque might be a performance bottleneck.
Our current implementation, see https://gitlab.anu.edu.au/mmtk/mmtk/blob/c453f6623bab8951df61cd19af760a44c0b50b8b/src/plan/semispace/sstracelocal.rs and https://gitlab.anu.edu.au/mmtk/mmtk/blob/c453f6623bab8951df61cd19af760a44c0b50b8b/src/util/global_pool.rs, follows the same pattern. We may or may not want to improve this, see https://gitlab.anu.edu.au/mmtk/mmtk/commit/f7fa69dfd3d69095ed32e4f6b19c7ab3d01968a7 commit message
As we will switch to work packets (https://github.com/mmtk/mmtk-core/issues/48), this issue seems irrelevant now.
In GitLab by @caizixian on Feb 10, 2018, 14:01
Yi's ISMM2016 paper shows that having a central controller thread to receive works and push them to the global deque might be a performance bottleneck.
Our current implementation, see https://gitlab.anu.edu.au/mmtk/mmtk/blob/c453f6623bab8951df61cd19af760a44c0b50b8b/src/plan/semispace/sstracelocal.rs and https://gitlab.anu.edu.au/mmtk/mmtk/blob/c453f6623bab8951df61cd19af760a44c0b50b8b/src/util/global_pool.rs, follows the same pattern. We may or may not want to improve this, see https://gitlab.anu.edu.au/mmtk/mmtk/commit/f7fa69dfd3d69095ed32e4f6b19c7ab3d01968a7 commit message