Lock-Free Work Stealing, Part 1: Basics

Re:Lock-Free Work Stealing, Part 1: Basics by audidude on Monday August 24, 2015 @07:56PM (#10113557)

I don't see any mention of where the author mitigates, or prevents the ABA problem. https://en.wikipedia.org/wiki/ABA_problem

- Some systems will use a double-CAS where one word is a counter.

- People lucky enough to use RCU can use grace-periods.

- You can also mitigate it (to a degree) using the bottom bits of the pointer and the kernel area of the pointer as a counter.

- Also, Hazard Pointers.

[ Parent | on HN ]

Re:Lock-Free Work Stealing, Part 1: Basics by mattnewport on Monday August 24, 2015 @11:19PM (#10114421)

This post doesn't go into his lock free work stealing queue implementation at all, he says that will be covered in a later post. I don't see how the ABA problem relates to any of the implementation details covered in this post?

[ Parent | on HN ]

Re:Lock-Free Work Stealing, Part 1: Basics by uxcn on Tuesday August 25, 2015 @01:21AM (#10114737)

Implementing a lock-free work stealing queue isn't difficult, especially with C/C++ atomics. I can't think of an implementation where ABA would be a problem.

[ Parent | on HN ]