Lock free fd lookup

Writer vs. writer starvation on NUMA is a lot harder. I don't know of any algorithm that handles lists with lots of concurrent updates and also scales well on large cpus, unless the underlying hardware is fair in its handling of exclusive cache lines.

Quoted 4 lines

Writer vs. writer starvation on NUMA is a lot harder. I don't know of any algorithm that handles lists with lots of concurrent updates and also scales well on large cpus, unless the underlying hardware is fair in its handling of exclusive cache lines.

How about MCS (Mellor-Crummey and Scott) locks?
Linux code:
http://oss.software.ibm.com/linux/patches/?patch_id=218
Something supposedly better:
http://user.it.uu.se/~zoranr/rh_lock/
Scott's list of 11 scalable synchronization algorithms:
http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/ss.html
Scott's collection of papers and so on:
http://www.cs.rochester.edu/u/scott/synchronization/
Simply asking Scott might be a wise move. He'd likely know of anything else that might fit the requirements. That's scott at cs.rochester.edu