Neil Horman

@kroeckx The current approach regarding refcounting is the latter you suggested. A refcount on the table itself isn't a valid approach, as the purpose of rcu locks is so that...

@kroeckx The former approach works now, and exactly the way you're describing it. Additions can be done with only a write side lock (not synchronizing on the read size), as...

@paulidale Thank you for that. Its an interesting approach. From my read of the paper, It sounds like the implementation is effectively a multi-layer hopscotch approach. i.e a key will...

@paulidale FWIW I found this example of iceberg: https://github.com/splatlab/iceberghashtable/tree/main And it does a number of atomic operations on shared data during a read. This may be a bad example of...

@paulidale something seems off with this iceberg implementation (which may not be suprising, given its just some random codebase on github). But its claiming in its test output that with...

Interesting data on the new hashtable with the other tests in the perf tools repo here: https://github.com/openssl/project/issues/440#issuecomment-1960074322

I'm not sure what happened here, but it says I closed and merged this PR, when I certainly didn't intend to, nor do I have permission to, and the commit...

New PR in https://github.com/openssl/openssl/pull/23671

I'm sorry, not sure I follow. Are you referring to the hashing algorithm? I chose a default of fnv1a, because its fast and fairly simple to implement, but it can...

@Sashan but #23887 didn't merge cleanly to 3.1/3.0 as per the comment [here ](https://github.com/openssl/openssl/pull/23887#issuecomment-2045856412), so there must be some minor line difference