Lock-free algorithms use atomic operations instead of locks to allow for thread-safe concurrent access to shared data structures. This avoids issues like deadlocks but makes implementation more complex. Common atomic operations include compare-and-swap (CAS) and load-linked/store-conditional (LL/SC). A key challenge is avoiding the ABA problem where a value is changed back to its original. Solutions involve adding version numbers or using more advanced operations like CAS2. Overall, lock-free approaches can improve performance over locking but require careful design to ensure correctness and avoid data corruption between threads.

1. Lock-Free Algorithms
Present by PanPan

2. Multi-Thread Programming Problem
a = a + 1
Read a
Calculate a + 1
Save a + 1
Thread A
Read a
Save a + 1
Calculate a + 1
Thread B
Read a
Save a + 1
Calculate a + 1
Time

3. Multi-Thread Programming Solution
●Blocking
○Mutex
○Critical Section
●Non-Blocking
○Wait-Free
○Lock-Free
○Obstruction-Free
●Local
○No Shared Data

4. Dead Lock
void FunctionA()
Enter Critical Section A
Leave Critical Section A
Leave Critical Section B
Enter Critical Section B
void FunctionB()
Enter Critical Section A
Leave Critical Section B
Leave Critical Section A
Enter Critical Section B
Do Something
Do Something
Do Something
Time
Call FunctionA
Return
(Or Close)
Do Something
Dead Lock
Dead Lock
Dead Lock
Dead Lock

5. What is Lock-Free?
A lock-free data structure is one that doesn't
use any mutex locks. The implication is that
multiple threads can access the data structure
concurrently without race conditions or data
corruption, even though there are no locks —
people would give you funny looks if you
suggested that std::list was a lock-free data
structure, even though it is unlikely that there
are any locks used in the implementation. [1]
Non-Blocking
Obstruction-Free
Lock-Free
Wait-Free
[1] http://www.justsoftwaresolutions.co.uk/threading/non_blocking_lock_free_and_wait_free.html

6. Advantages and Disadvantages
●No "Lock" (Also No Dead Lock)
●Better Performance
●Hard to Implement

7. Main Concept
●Use Atom Operation to Modify
Critical Data
●Enter a Spin State When Atom
Operation Fail
Atom Operation
Success?
Finish
No
Yes

8. Compare and Swap
bool CAS(uint32* pointer, uint32 old_value, uint32 new_value)
{
if (*pointer == old_value)
{
*pointer = new_value;
return true;
}
return false;
}
InterlockedCompareExchange - Windows API

9. Other Atom Operation
●CAS2 - Compare and Swap 2
●CASN - Compare and Swap N
●DCAS - Double Compare Swap
●MCAS - Multiword Compare and Swap
●LL/SC - Load Linked / Store Conditional

10. Example - Stack
struct Node
{
volatile Node* next;
};
class Stack
{
volatile Node* head;
public:
void Push(Node* node);
void Pop();
Stack() : head(0) {}
};
void Stack::Push(Node* node)
{
while (true)
{
node->next = head;
if (CAS(&head, node->next, node))
break;
}
}
void Stack::Pop()
{
while (true)
{
node* old_head = head;
if (head == 0)
break;
node* next_node = old_head->next;
if (CAS(&head, old_head, next_node))
break;
}
}
Something Wrong

11. ABA Problem
Thread A
A
B
C
A
B
C
Head
Pop
Thread B
A
B
C
Pop
C
A
Push
Head
Head
Head
node* old_head = head;
if (head == 0)
break;
node* next_node = old_head->next;
if (CAS(&head, old_head, next_node))
break;

12. Solute ABA Problem
●LL/SC
○It would return false when target have
done a "write" action
●CAS2
○Add a versioned value

13. Example - Stack(fix Pop)
void Stack::Pop()
{
while (true)
{
node* old_head = head;
int old_version = version;
if (head == 0)
break;
node* next_node = old_head->next;
if (CAS2(&head, old_head, old_version, next_node, old_version + 1))
break;
}
}
class Stack
{
volatile Node* head;
volatile int version;
public:
void Push(Node* node);
void Pop();
Stack() : head(0), version(0) {}
};
Add Version
Check Version

14. Summary
●Lock-Free has Better Performance Than Blocking
●Lock-Free has no "Lock"
●Lock-Free is Hard To Implement
●Lock-Free's Spirit - Atom Operation
●Lock-Free's Trap - ABA Problem

15. Notes and References
●Toby Jones - Lock-Free Algorithms(Game Programming
Gems 6)
●Jean-Francois Dube - Efficient and Scalable Multi-Core
Programming(Game Programming Gems 8)
●Zhou Wei Ming - Multi-core Computing and Programming

16. Any Questions?

17. Thanks for your attention
By PanPan
2010-06-29