On: Lock-free multithreading with atomic operations

[jchw]

Note that lock-free is not always faster because contention is complicated.

[vardump → jchw]

Also note that lock-free can scale worse, again due to contention.

If you want scalability, you want to minimize inter-core and inter-socket traffic.

[jchw → vardump]

This is the more important point, realistically. I think I probably would've been better off saying 'lock-free is not a panacea.'

All in all, scaling up contentious operations is hard.