Skip to content

Latest commit

 

History

History
71 lines (41 loc) · 1.63 KB

README.md

File metadata and controls

71 lines (41 loc) · 1.63 KB

Wait-free simulation for a Lock-free algorithm.

Implementation of a wait-free simulation for a lock-free algorithm as described in the paper - A Practical Wait-Free Simulation for Lock-Free Data Structures.

Components of the algorithm

  1. Normalized Lock Free Form
  2. Wait-free queue
  3. Contention Measure
  4. Cas Descriptor
  5. Wait-Free simulator

The ABA problem

This section maybe a little convoluted. It is a WIP.

Say, the head of a linked list points to a node A

head -> A (@0x1)

We now want to insert node B.

insert B (@0x2); B.next = 0x1

We do this using a compare_and_swap primitive.

CAS(head, 0x1, 0x2) -> Succeed if A is still at the head -> fail if A is no longer at the head (maybe due to the insertion from another thread), because B.next would need to be different.

However, there is a possibilty of a problem.

Imagine someone else meanwhile

insert C (@0x3); C.next = 0x1

CAS(head, 0x1, 0x3)

Now lets imagine someone removes A

remove A

CAS(C.next, 0x1, 0x0)

Now if someone inserts a node D

insert D (@0x1); and D.next = 0x3

C is at the head

CAS(head, 0x3, 0x1)

So the final LL

head -> D -> C -> .

and now ...

insert of B continues.

CAS(head, 0x1, 0x2) -> this CAS should have failed ideally

head -> B -> A(@0x1) actually D(@0x1) -> C(@0x3)

Even though the last CAS happened to succeed since the values were same, but the thing we are pointing aren't really the same.

This is known as the ABA problem.

References

  1. Full length Paper
  2. Jon Gjengset's Stream