Nicety of Java 8 Multithreading

Nicety of Java 8 Multithreading

• 1.
  1 Nicety of java8 multithreading for advanced Maksym Voroniy Software Architect of GlobalLogic
• 2.
  2 Common points01
• 3.
  3 History Text The Dream Timesharing Stone Age Intel 80386 Multiple data multiple handlers Multiple processo rs Multiple data single handler Hyper threading IBM 360 CUDA Multicore Today
• 4.
  4 Point #1 Maurice Herlihy& Nir Shavit “The Art of Multiprocessor Programming”
• 5.
  5 Atomicity vs Consistence staticint a = 5; a *= 3; int b = 22; b += a; mov EAX, 5 mov ds:[a], EAX mul 3 mov ds:[a], EAX mov EBX, 22 add EBX, EAX;vs add EBX,ds:[a] atomicity consistence
• 6.
  6 Java 8 FAAvs CAS02
• 7.
  7 Nicety #1 -Java 8 Atomics http://ashkrit.blogspot.com/2014/02/atomicinteger-java-7-vs-java-8.html fetch-and-add vs compare-and-swap (CAS) JDK 8 - AtomicInteger public final int getAndIncrement() { return unsafe.getAndAddInt( this, valueOffset, 1); } JDK7 - AtomicInteger public final int getAndIncrement() { for (;;) { int current = get(); int next = current + 1; if (compareAndSet(current, next)) return current; } }
• 8.
  8 Why Atomic important? Text ConcurrentHashMap<> Collections.synchronizedMap(newHashMap<>()) • ConcurrentLinkedDeque/Queue • ConcurrentSkipListSet/Map • ConcurrentHashMap • wait-free, if every concurrent operation is guaranteed to be finished in a finite number of steps • lock-free, if some concurrent operations are guaranteed to be finished in a finite number of steps. DARE develop concurrent hash-map TODAY tomorrow
• 9.
  9 ConcurrentHashMap StampedLock goes tostage Text Simple cache implementation (value by key): (T key, Map<T, V> map) -> { return map.get(i); } synchronizedMap(new HashMap<>()) (T key, Map<T, V> map) -> { synchronized (map){ return map.get(i); } } HashMap + ReadWriteLock (T key, Map<T, V> map) -> { Lock l = rwLock.readLock(); l.lock(); try { return map.get(i); } finally { l.unlock(); }}
• 10.
  10 StampedLock goes tostage (II) Text HashMap + StampedLock (T key, Map<T, V> map) -> { long l = stampedLock.readLock(); try { return map.get(i); } finally { stampedLock.unlockRead(l); }} HashMap + StampedLock+Optimistic (T key, Map<T, V> map) -> { long stamp = optimisticLock.tryOptimisticRead(); Integer r = map.get(i); if (!optimisticLock.validate(stamp)) { stamp = optimisticLock.readLock(); try { return map.get(i); } finally { optimisticLock.unlockRead(stamp); } } return r; } 200`000 op for 10 Threads Nanoseconds ConcurrentHashMap 72645.057 Optimistic 359819.982 StampedLock 1303853.525 ReadWriteLock 1326939.766 Locked 1922399.053
• 11.
  11 Java 8 Parallelism03
• 12.
  12 .parallelStream() Text //from Oracle tutorial: doubleaverage = roster .parallelStream() .filter(p -> p.getGender() == Person.Sex.MALE) .mapToInt(Person::getAge) .average() .getAsDouble() • Java runtime partitions the stream into multiple substreams. Aggregate operations iterate over and process these substreams in parallel and then combine the results static ExecutorService newWorkStealingPool() static ExecutorService newFixedThreadPool(int nThreads) static ScheduledExecutorService newScheduledThreadPool(int corePoolSize) static ExecutorService newCachedThreadPool(ThreadFactory threadFactory) • Work Stealing (An idle thread steals work from a thread having tasks queued up more than it can process currently) • Ability to recursively decompose the tasks and collect the results. (Apparently, this requirement must have popped up along with the conception of the notion of parallel processing... but lacked a solid implementation framework in Java till Java 7) ForkJoinPool
• 13.
  13 The trap! List<SomeClass> list= // A list of objects list.parallelStream() .map(this::veryLongProcessing) .collect(toList()); partition Stream-1 Stream-2 Stream-3 Core -1 Core -2 ForkJoinPool forkJoinPool = new ForkJoinPool(2); forkJoinPool.submit(() -> list.parallel() .map(this::veryLongProcessing) .collect(toList()) ).get(); "Arranges to asynchronously execute this task in the pool the current task is running in, if applicable, or using the ForkJoinPool.commonPool() if not in ForkJoinPool()"
• 14.
  14 The Future Text Time sharing StoneAge Intel 80386 Hyper threading IBM 360 CUDA Multicore HTMSTM
• 15.
  15 Future direction • HowDatabases works (SQL& NoSQL) • Quantum computing • Advanced data structures (R-Tree, B-Tree, Trie, Skip-lists, Heaps…) http://goo.gl/forms/9LauaeJBlO N
• 16.
  16 Thank you! Any Questions?