Use cases for weakCAS

В java.util.concurrent.atomic.AtomicXXX есть волшебный метод weakCompareAndSet, который отличается от оригинального  compareAndSet двумя моментами:

1) может  spuriously fail "The weak version may be more efficient in the normal case, but differs in that any given invocation of weakCompareAndSetmethod may fail, even spuriously (that is, for no apparent reason)."

2) не устанавливает happend-before отношение "weakCompareAndSet atomically reads and conditionally writes a variable, is ordered with respect to other memory operations on that variable, but otherwise acts as an ordinary non-volatile memory operation."

---

Для такого поведения (второй пункт) найдено пока два Use Case:

A) счетчики

B) Michael-Scott queues (надо попробовать реализовать)

P.S. Как понимаю, j.u.c.ConcurrentLinkedQueue реализовано на основе Michael-Scott queue, но использует "strong" CAS.

Cliff Click vs Doug Lea: about fence-less CAS

Interesting conversation(s) with Doug Lea

Abstract
We are discussing the merits of a no-fence CAS instruction - and exposing this instruction at the Java level. Since Intel does not have a no-fence this whole idea is a no-op for Intel. Right now there is a weak spurious-fail-allowed CAS in the Unsafe to mimic load-linked/store-conditional ops, to get buy-in from IBM whose chips do LL/SC instead of CAS. Doug wants to the fence-less CAS to allow spurious failure, and I do not because it causes me to add retry loops in places.

Dr. Cliff Click: Towards a Scalable Non-Blocking Coding Style

Rapporteur at JavaOne 2008 from Dr. Cliff Click from Azul Systems one of the Best in the world specialist in Super-Scalable(768 and Up cores) Non-Blocking structures for Java.

Author of high-scale-lib (free lib at SourceForge) (lib has BitVector, HashTable and (FIFO-Queue in progress now)).

Dr. Cliff Click talk how by using FSM(Finite State Machine) and atomic-CAS create scalable non-locking structures.

"Classic reader/writer lock chokes w/ >100 CPUs

• Contention on single reader-count word limits scaling"

"• Lock-Free: Each CAS makes progress

• CAS success is local progress

• CAS failure means another CAS succeeded(global progress, local starvation)"

"JDK API mistake: witness turned into a boolean

• Hence failure-for-cause can not be distinguished from spurious-failure"

"Highly scalable (proven scalable to ~1000 CPUs)"

P.S. FSM model for verification maybe interesting for you, Igor?