Spinlocks and Barrier Synchronization

Question 1

Spinlock

Answer 1

• Each java object has an intrinsic lock
• So far we have dealt with threads that block
  awaiting access to a shared resource
  (This is implicit using synchronized keyword)
• An alternative is to keep the thread active and continuously ‘spinning’ attempting to acquire the lock
   This is a spin lock
   (Potentially) improves threading performance and CPU usage

Question 2

An Attempted Spinlock Implementation\

Answer 2

Question 3

Spinlock execution

Answer 3

Question 4

Spinlock Solution

Answer 4

Question 5

Disabling Interrupts

Answer 5

• Pre-emptive context switch only happens when interrupt occurs
• Could disable interrupts to prevent context switch in critical section

While this works, comes with many disadvantages…
1. Interrupts might be disabled frequently and for a long
time
 Clock ticks, I/O events could be missed
2. Will not be sufficient when you have more than one processor
* More than once thread could be running concurrently
3. Error proneness
 Forgetting to call release_lock() means interrupts disabled forever

Question 6

Machine Instruction

Answer 6

test-and-set, comp-and-swap, fetch-and-add

• Atomic machine instruction
  
  • Sets variable passed to true,
  • Tells if variable was true or false before being set to true

Question 7

Machine Instruction: test_and_set() lock

Answer 7

Question 8

Software Only – Peterson’s Algorithm

Answer 8

Assumes atomic reads and writes
- Only works with two threads (can be generalized to n threads)
- Assumes thread ID are 0 and 1

Question 9

In green and pink are instructions from
processes 0 and 1, respectively

Question 10

Blocking vs spinning locks

Answer 10

Blocking:
- Scheduler blocks threads while they wait
- Good for long critical sections
- Frequent queue management if locks
accessed frequently

Spinning:
- Sit in a tight loop until lock acquisition
- Good for short critical sections
- Avoid queue management