Lecture 16: Concurrent Linked Lists

Overview

  1. Lab 04 Tips
  2. Set as a Linked List
  3. Synchronization Philosophies
  4. Coarse-grained Linked List
  5. Fine-grained Linked List
  6. Optimistic Linked List

Lab 04 Tips

Prime Tasks

Suggested structure:

  1. Compute small primes (up to Math.sqrt(MAX))
    • use Primes.getPrimesUpTo(int max)
  2. Use small primes and SoE to find rest of primes
    • consider a block [n, n + k] of numbers from n to n + k
    • for each small prime p, remove multiples of p from [n, n + k]
    • remaining numbers are prime
  3. Write primes from step 2 in order to primes array

Have a dedicated thread for step 3; use several threads to perform step 2 in parallel

Prime Tasks in Pictures

Performance Optimizations

  1. Use reasonably small objects
    • several smaller tasks/arrays are better than few larger ones
    • arrays should fit in low-level cache
  2. Minimize sharing of objects
    • each object should be written by one thread, and read by at most one other thread
    • think of “passing” completed tasks from producer to consumer
    • limit concurrent access

A Set

A Set of elements:

  • store a collection of distinct elements
  • add an element
    • no effect if element already there
  • remove an element
    • no effect if not present
  • check if set contains an element

An Interface

public interface SimpleSet<T> {
    /*
     * Add an element to the SimpleSet. Returns true if the element
     * was not already in the set.
     */
    boolean add(T x);

    /*
     * Remove an element from the SimpleSet. Returns true if the
     * element was previously in the set.
     */
    boolean remove(T x);

    /*
     * Test if a given element is contained in the set.
     */
    boolean contains(T x);
}

Implementing SimpleSet

Store elements in a linked list of nodes

  • Each Node stores:
    • reference to the stored object
    • reference to the next Node
    • a key associated with the object
      • use hash code of object
      • keys can be sorted
  • The list stores
    • reference to head node
    • a tail node
    • head and tail have min and max key values

Our Goals

  1. Correctness, safety, liveness
    • linearizability
    • deadlock-freedom
    • starvation-freedom?
    • nonblocking???
  2. Performance
    • parallelism?

Synchronization Philosophies

Synchronization Philosophies

  1. Coarse-Grained
    • lock whole data structure for every operation
  2. Fine-Grained
    • only lock what is needed to avoid disaster
  3. Optimisitc
    • don’t lock anything to read
    • lock to modify
  4. Lazy
    • use “logical” removal
    • only use locks occasionally
  5. Nonblocking
    • use atomics, not locks!

Coarse-grained Linked List

Coarse-grained Linked List

One lock for whole data structure

For any operation:

  1. Lock entire list
  2. Perform operation
  3. Unlock list

Coarse-grained List Figure

Add item with Key 5

Step 1: Lock List

Step 2: Find Correct Location

Step 3: Do Insertion

Step 4: Unlock List

In Code

  • Look at implementation
  • Test performance
  • This is our baseline!

Fine-grained Linked List

Fine-grained Linked List

One lock per node

For any operation:

  1. Store curr, pred (initialized to head)
    • lock pred
  2. Hand-over-hand locking:
    • lock curr = pred.next
    • unlock pred
    • set pred = curr
    • repeat until correct location found
  3. Perform operation
  4. Release locks

Fine-grained List Figure

Add item with Key 5

Step 1: Set and Lock pred

Step 2: Hand over Hand Locking

Step 3: Perform Insert

Step 4: Release Locks

Test

  • More efficient than coarse?
  • Why or why not?
  • What affects performance?

Optimistic Linked List

Optimistic Linked List

One lock per node

For any operation:

  1. Find location
  2. Acquire locks
  3. Validate location
    • go back to 1 if this fails!
  4. Perform operation
  5. Release locks

Optimistic List Figure

Add item with Key 5

Step 1: Find Location

Step 2: Acquire Locks

Step 3: Validate

Step 4: Perform Operation

Step 5: Release Locks

Why Could Validation fail?

Questions

  • Liveness: Deadlock-free? Starvation-free?
  • Performance: When do you expect good performance? When not?

Next Time:

  1. Lazy Synchronized List
  2. Nonblocking Synchronized List
  3. Other Data Structures