Develop Reference

SpinWait vs Sleep waiting. Which one to use?

Is it efficient to
SpinWait.SpinUntil(() => myPredicate(), 10000)
for a timeout of 10000ms
or
Is it more efficient to use Thread.Sleep polling for the same condition
For example something along the lines of the following SleepWait function:
public bool SleepWait(int timeOut)
{
Stopwatch stopwatch = new Stopwatch();
stopwatch.Start();
while (!myPredicate() && stopwatch.ElapsedMilliseconds < timeOut)
{
Thread.Sleep(50)
}
return myPredicate()
}
I'm concerned that all the yielding of SpinWait may not be a good usage pattern if we are talking about timeouts over 1sec? Is this a valid assumption?
Which approach do you prefer and why? Is there another even better approach?
Update - Becoming more specific:
Is there a way to Make BlockingCollection Pulse a sleeping thread when it reaches bounded capacity? I rather avoid a busy waits alltogether as Marc Gravel suggests.

In .NET 4 SpinWait performs CPU-intensive spinning for 10 iterations before yielding. But it does not return to the caller immediately after each of those cycles; instead, it calls Thread.SpinWait to spin via the CLR (essentially the OS) for a set time period. This time period is initially a few tens of nano-seconds but doubles with each iteration until the 10 iterations are complete. This enables clarity/predictability in the total time spent spinning (CPU-intensive) phase, which the system can tune according to conditions (number of cores etc.). If SpinWait remains in the spin-yielding phase for too long it will periodically sleep to allow other threads to proceed (see J. Albahari's blog for more information). This process is guaranteed to keep a core busy...
So, SpinWait limits the CPU-intensive spinning to a set number of iterations, after which it yields its time slice on every spin (by actually calling Thread.Yield and Thread.Sleep), lowering its resource consumption. It will also detect if the user is running a single core machine and yield on every cycle if that is the case.
With Thread.Sleep the thread is blocked. But this process will not be as expensive as the above in terms of CPU.

The best approach is to have some mechanism to actively detect the thing becoming true (rather than passively polling for it having become true); this could be any kind of wait-handle, or maybe a Task with Wait, or maybe an event that you can subscribe to to unstick yourself. Of course, if you do that kind of "wait until something happens", that is still not as efficient as simply having the next bit of work done as a callback, meaning: you don't need to use a thread to wait. Task has ContinueWith for this, or you can just do the work in an event when it gets fired. The event is probably the simplest approach, depending on the context. Task, however, already provides most-everything you are talking about here, including both "wait with timeout" and "callback" mechanisms.
And yes, spinning for 10 seconds is not great. If you want to use something like your current code, and if you have reason to expect a short delay, but need to allow for a longer one - maybe SpinWait for (say) 20ms, and use Sleep for the rest?
Re the comment; here's how I'd hook an "is it full" mechanism:
private readonly object syncLock = new object();
public bool WaitUntilFull(int timeout) {
if(CollectionIsFull) return true; // I'm assuming we can call this safely
lock(syncLock) {
if(CollectionIsFull) return true;
return Monitor.Wait(syncLock, timeout);
}
}
with, in the "put back into the collection" code:
if(CollectionIsFull) {
lock(syncLock) {
if(CollectionIsFull) { // double-check with the lock
Monitor.PulseAll(syncLock);
}
}
}

Monitor.Wait - while or if?

Currently, I'm learning for a multithreading exam. I read the good threading article of albahari. I've got a question at the monitor usage - why is here used a loop in place of an if?
lock (_locker)
{
while (!_go) //why while and not if?
Monitor.Wait (_locker); // _lock is released
// lock is regained
...
}
I think, that an if would be sufficient.
I'm afraid, that I don't understand the article completely.
//Edit
Example-Code:
class SimpleWaitPulse
{
static readonly object _locker = new object();
static bool _go;
static void Main()
{ // The new thread will block
new Thread (Work).Start(); // because _go==false.
Console.ReadLine(); // Wait for user to hit Enter
lock (_locker) // Let's now wake up the thread by
{ // setting _go=true and pulsing.
_go = true;
Monitor.Pulse (_locker);
}
}
static void Work()
{
lock (_locker)
while (!_go)
Monitor.Wait (_locker); // Lock is released while we’re waiting
Console.WriteLine ("Woken!!!");
}
}

It just depends on the situation. In this case the code is just waiting for _go to be true.
Every time _locker is pulsed it will check to see if _go has been set to true. If _go is still false, it will wait for the next pulse.
If an if was used instead of a while, it would only wait once (or not at all if _go was already true), and would then continue on after a pulse, regardless of the new state of _go.
So how you use Monitor.Wait() depends entirely on your specific needs.

It really just depends on the situation. But first, we need to clarify how Monitors work. When a thread proceeds to signal a thread through Monitor.Pulse(), there is usually no guarantee that the signaled thread will actually run next. This means that it is possible for other threads to run before the signaled thread and change the condition under which it was okay for the signaled thread to proceed. This means that the signaled thread still needs to check if is safe for it to proceed after being woken up (ie the while loop). However, certain rare synchronization problems allow you to make the assumption that once a thread has been signaled to wake up (ie Monitor.Pulse()), no other thread has the ability to change the condition under which it is safe to proceed (ie. the if condition).

I wrote an article that might help here: Wait and Pulse demystified
There's more going on than is immediately obvious.

I've got a question at the monitor usage - why is here used a loop in
place of an if?
There is a well known rule when working with Pulse and Wait that states that when in doubt prefer while over an if. Clearly, either one will work in this case, but in almost every other situation while is required. In fact, there are very few (if any) scenarios where using a while loop would produce an incorrect result. That is the basis for this general rule. The author used a while loop because he was trying to stick with the tried-and-true pattern. He even provides the template in the same article. Here is it is:
lock (_locker)
while ( <blocking-condition> )
Monitor.Wait (_locker);

The simplest way to write correct code with Monitor.Wait is to assume the system will regard it as "advisory", and assume that the system may arbitrarily wake any waiting thread any time it can acquire the lock, without regard for whether Pulse has been called. The system usually won't do so, of course, but if a program is using Wait and Pulse properly, its correctness should not be affected by having Wait calls arbitrarily exit early for no reason. Essentially, one should regard Wait as a means of telling the system "Continuing execution past here will be a waste of time unless or until someone else calls Pulse".

C# alternative to lock if volatile isn't a good idea

I'm sorry I know this topic has been done to death (I've read I've read this and this and a few more) but there is one issue I have which I am not sure how to do 'correctly'.
Currently my code for a multithreaded Sudoku strategy is the following:
public class MultithreadedStrategy : ISudokuSolverStrategy
{
private Sudoku Sudoku;
private List<Thread> ThreadList = new List<Thread>();
private Object solvedLocker = new Object();
private bool _solved;
public bool Solved // This is slow!
{
get
{
lock (solvedLocker)
{
return _solved;
}
}
set
{
lock (solvedLocker)
{
_solved = value;
}
}
}
private int threads;
private ConcurrentQueue<Node> queue = new ConcurrentQueue<Node>();
public MultithreadedStrategy(int t)
{
threads = t;
Solved = false;
}
public Sudoku Solve(Sudoku sudoku)
{
// It seems concevable to me that there may not be
// a starting point where there is only one option.
// Therefore we may need to search multiple trees.
Console.WriteLine("WARNING: This may require a large amount of memory.");
Sudoku = sudoku;
//Throw nodes on queue
int firstPos = Sudoku.FindZero();
foreach (int i in Sudoku.AvailableNumbers(firstPos))
{
Sudoku.Values[firstPos] = i;
queue.Enqueue(new Node(firstPos, i, false, Sudoku));
}
//Setup threads
for (int i = 0; i < threads; i++)
{
ThreadList.Add(new Thread(new ThreadStart(ProcessQueue)));
ThreadList[i].Name = String.Format("Thread {0}", i + 1);
}
//Set them running
foreach (Thread t in ThreadList)
t.Start();
//Wait until solution found (conditional timeout?)
foreach (Thread t in ThreadList)
t.Join();
//Return Sudoku
return Sudoku;
}
public void ProcessQueue()
{
Console.WriteLine("{0} running...",Thread.CurrentThread.Name);
Node currentNode;
while (!Solved) // ACCESSING Solved IS SLOW FIX ME!
{
if (queue.TryDequeue(out currentNode))
{
currentNode.GenerateChildrenAndRecordSudoku();
foreach (Node child in currentNode.Children)
{
queue.Enqueue(child);
}
// Only 1 thread will have the solution (no?)
// so no need to be careful about locking
if (currentNode.CurrentSudoku.Complete())
{
Sudoku = currentNode.CurrentSudoku;
Solved = true;
}
}
}
}
}
(Yes I have done DFS with and without recursion and using a BFS which is what the above strategy modifies)
I was wondering whether I am allowed to change my private bool _solved; to a private volatile solved; and get rid of the accessors. I think this might be a bad thing because my ProcessQueue() method changes the state of _solved Am I correct? I know booleans are atomic but I don't want compiler optimisations to mess up the order of my read/write statements (esp. since the write only happens once).
Basically the lock statement adds tens of seconds to the run time of this strategy. Without the lock it runs an awful lot faster (although is relatively slow compared to a DFS because of the memory allocation within currentNode.GenerateChildrenAndRecordSudoku()

Before getting into alternatives: it is probably safe to go with a low-lock solution here by making access to the boolean volatile. This situation is ideal, as it is unlikely that you have complex observation-ordering requirements. ("volatile" does not guarantee that multiple volatile operations are observed to have consistent ordering from multiple threads, only that reads and writes have acquire and release semantics.)
However, low-lock solutions make me very nervous and I would not use one unless I was sure I had need to.
The first thing I would do is find out why there is so much contention on the lock. An uncontended lock should take 20-80 nanoseconds; you should only get a significant performance decrease if the lock is contended. Why is the lock so heavily contended? Fix that problem and your performance problems will go away.
The second thing I might do if contention cannot be reduced is to use a reader-writer lock. If I understand your scenario correctly, you will have many readers and only one writer, which is ideal for a reader-writer lock.
Leaving the question of volatility aside: as others have pointed out, there are basic mistakes in your threading logic like spinning on a boolean. This stuff is hard to get right. You might consider using the Task Parallel Library here as a higher-level abstraction than rolling your own threading logic. The TPL is ideally suited for problems where significant work must be done on multiple threads. (Note that the TPL does not magically make not-thread-safe code thread-safe. But it does provide a higher level of abstraction, so that you are dealing with Tasks rather than Threads. Let the TPL schedule the threads for you.)
Finally: the idea that a sudoku solver should take tens of seconds indicates to me that the solver is, frankly, not very good. The sudoku problem is, in its theoretically worst possible case, a hard problem to solve quickly no matter how many threads you throw at it. But for "newspaper" quality sudokus you should be able to write a solver that runs in a fraction of a second. There's no need to farm the work out to multiple threads if you can do the whole thing in a few hundred milliseconds.
If you're interested, I have a C# program that quickly finds solutions to sudoku problems here:
http://blogs.msdn.com/b/ericlippert/archive/tags/graph+colouring/

So the first thing, fix you're while loop to just join the threads...
//Set them running
foreach (Thread t in ThreadList)
t.Start();
//Wait until solution found (conditional timeout?)
foreach (Thread t in ThreadList)
t.Join(/* timeout optional here */);
Then there is issue with when to shutdown the threads. My advise is to introduce a wait handle on the class and then in the worker threads just loop on that...
ManualResetEvent mreStop = new ManualResetEvent(false);
//...
while(!mreStop.WaitOne(0))
{
//...
Now just modify the Solved property to signal all threads that they should quit...
public bool Solved
{
get
{
return _solved;
}
}
// As Eric suggests, this should be a private method, not a property set.
private void SetCompleted()
{
_solved = value;
mreStop.Set();
}
The benefit to this approach is that if a thread fails to quit within a timeout period you can signal the mreStop to stop the workers without setting _solved to true.

volatile IS used to prevent optimizations such as caching and reordering of reads/writes for a single variable. Using it in this case is exactly what it's designed for. I don't see what your concern is.
lock is a slow yet working alternative because it introduces a memory fence implicitly, but in your case you are using a lock just for the memory fence side-effect, which is not really a nice idea.

does Monitor.Wait Needs synchronization?

I have developed a generic producer-consumer queue which pulses by Monitor in the following way:
the enqueue :
public void EnqueueTask(T task)
{
_workerQueue.Enqueue(task);
Monitor.Pulse(_locker);
}
the dequeue:
private T Dequeue()
{
T dequeueItem;
if (_workerQueue.Count > 0)
{
_workerQueue.TryDequeue(out dequeueItem);
if(dequeueItem!=null)
return dequeueItem;
}
while (_workerQueue.Count == 0)
{
Monitor.Wait(_locker);
}
_workerQueue.TryDequeue(out dequeueItem);
return dequeueItem;
}
the wait section produces the following SynchronizationLockException :
"object synchronization method was called from an unsynchronized block of code"
do i need to synch it? why ? Is it better to use ManualResetEvents or the Slim version of .NET 4.0?

Yes, the current thread needs to "own" the monitor in order to call either Wait or Pulse, as documented. (So you'll need to lock for Pulse as well.) I don't know the details for why it's required, but it's the same in Java. I've usually found I'd want to do that anyway though, to make the calling code clean.
Note that Wait releases the monitor itself, then waits for the Pulse, then reacquires the monitor before returning.
As for using ManualResetEvent or AutoResetEvent instead - you could, but personally I prefer using the Monitor methods unless I need some of the other features of wait handles (such as atomically waiting for any/all of multiple handles).

From the MSDN description of Monitor.Wait():
Releases the lock on an object and blocks the current thread until it reacquires the lock.
The 'releases the lock' part is the problem, the object isn't locked. You are treating the _locker object as though it is a WaitHandle. Doing your own locking design that's provably correct is a form of black magic that's best left to our medicine man, Jeffrey Richter and Joe Duffy. But I'll give this one a shot:
public class BlockingQueue<T> {
private Queue<T> queue = new Queue<T>();
public void Enqueue(T obj) {
lock (queue) {
queue.Enqueue(obj);
Monitor.Pulse(queue);
}
}
public T Dequeue() {
T obj;
lock (queue) {
while (queue.Count == 0) {
Monitor.Wait(queue);
}
obj = queue.Dequeue();
}
return obj;
}
}
In most any practical producer/consumer scenario you will want to throttle the producer so it cannot fill the queue unbounded. Check Duffy's BoundedBuffer design for an example. If you can afford to move to .NET 4.0 then you definitely want to take advantage of its ConcurrentQueue class, it has lots more black magic with low-overhead locking and spin-waiting.

The proper way to view Monitor.Wait and Monitor.Pulse/PulseAll is not as providing a means of waiting, but rather (for Wait) as a means of letting the system know that the code is in a waiting loop which can't exit until something of interest changes, and (for Pulse/PulseAll) as a means of letting the system know that code has just changed something that might cause satisfy the exit condition some other thread's waiting loop. One should be able to replace all occurrences of Wait with Sleep(0) and still have code work correctly (even if much less efficiently, as a result of spending CPU time repeatedly testing conditions that haven't changed).
For this mechanism to work, it is necessary to avoid the possibility of the following sequence:
The code in the wait loop tests the condition when it isn't satisfied.
The code in another thread changes the condition so that it is satisfied.
The code in that other thread pulses the lock (which nobody is yet waiting on).
The code in the wait loop performs a Wait since its condition wasn't satisfied.
The Wait method requires that the waiting thread have a lock, since that's the only way it can be sure that the condition it's waiting upon won't change between the time it's tested and the time the code performs the Wait. The Pulse method requires a lock because that's the only way it can be sure that if another thread has "committed" itself to performing a Wait, the Pulse won't occur until after the other thread actually does so. Note that using Wait within a lock doesn't guarantee that it's being used correctly, but there's no way that using Wait outside a lock could possibly be correct.
The Wait/Pulse design actually works reasonably well if both sides cooperate. The biggest weaknesses of the design, IMHO, are (1) there's no mechanism for a thread to wait until any of a number of objects is pulsed; (2) even if one is "shutting down" an object such that all future wait loops should exit immediately (probably by checking an exit flag), the only way to ensure that any Wait to which a thread has committed itself will get a Pulse is to acquire the lock, possibly waiting indefinitely for it to become available.

C# Threading and Queues

This isn't about the different methods I could or should be using to utilize the queues in the best manner, rather something I have seen happening that makes no sense to me.
void Runner() {
// member variable
queue = Queue.Synchronized(new Queue());
while (true) {
if (0 < queue.Count) {
queue.Dequeue();
}
}
}
This is run in a single thread:
var t = new Thread(Runner);
t.IsBackground = true;
t.Start();
Other events are "Enqueue"ing else where. What I've seen happen is over a period of time, the Dequeue will actually throw InvalidOperationException, queue empty. This should be impossible seeing as how the count guarantees there is something there, and I'm positive that nothing else is "Dequeue"ing.
The question(s):
Is it possible that the Enqueue actually increases the count before the item is fully on the queue (whatever that means...)?
Is it possible that the thread is somehow restarting (expiring, reseting...) at the Dequeue statement, but immediately after it already removed an item?
Edit (clarification):
These code pieces are part of a Wrapper class that implements the background helper thread. The Dequeue here is the only Dequeue, and all Enqueue/Dequeue are on the Synchronized member variable (queue).

Using Reflector, you can see that no, the count does not get increased until after the item is added.
As Ben points out, it does seem as you do have multiple people calling dequeue.
You say you are positive that nothing else is calling dequeue. Is that because you only have the one thread calling dequeue? Is dequeue called anywhere else at all?
EDIT:
I wrote a little sample code, but could not get the problem to reproduce. It just kept running and running without any exceptions.
How long was it running before you got errors? Maybe you can share a bit more of the code.
class Program
{
static Queue q = Queue.Synchronized(new Queue());
static bool running = true;
static void Main()
{
Thread producer1 = new Thread(() =>
{
while (running)
{
q.Enqueue(Guid.NewGuid());
Thread.Sleep(100);
}
});
Thread producer2 = new Thread(() =>
{
while (running)
{
q.Enqueue(Guid.NewGuid());
Thread.Sleep(25);
}
});
Thread consumer = new Thread(() =>
{
while (running)
{
if (q.Count > 0)
{
Guid g = (Guid)q.Dequeue();
Console.Write(g.ToString() + " ");
}
else
{
Console.Write(" . ");
}
Thread.Sleep(1);
}
});
consumer.IsBackground = true;
consumer.Start();
producer1.Start();
producer2.Start();
Console.ReadLine();
running = false;
}
}

Here is what I think the problematic sequence is:
(0 < queue.Count) evaluates to true, the queue is not empty.
This thread gets preempted and another thread runs.
The other thread removes an item from the queue, emptying it.
This thread resumes execution, but is now within the if block, and attempts to dequeue an empty list.
However, you say nothing else is dequeuing...
Try outputting the count inside the if block. If you see the count jump numbers downwards, someone else is dequeuing.

Here's a possible answer from the MSDN page on this topic:
Enumerating through a collection is
intrinsically not a thread-safe
procedure. Even when a collection is
synchronized, other threads can still
modify the collection, which causes
the enumerator to throw an exception.
To guarantee thread safety during
enumeration, you can either lock the
collection during the entire
enumeration or catch the exceptions
resulting from changes made by other
threads.
My guess is that you're correct - at some point, there's a race condition happening, and you end up dequeuing something that isn't there.
A Mutex or Monitor.Lock is probably appropriate here.
Good luck!

Are the other areas that are "Enqueuing" data also using the same synchronized queue object? In order for the Queue.Synchronized to be thread-safe, all Enqueue and Dequeue operations must use the same synchronized queue object.
From MSDN:
To guarantee the thread safety of the
Queue, all operations must be done
through this wrapper only.
Edited:
If you are looping over many items that involve heavy computation or if you are using a long-term thread loop (communications, etc.), you should consider having a wait function such as System.Threading.Thread.Sleep, System.Threading.WaitHandle.WaitOne, System.Threading.WaitHandle.WaitAll, or System.Threading.WaitHandle.WaitAny in the loop, otherwise it might kill system performance.

question 1: If you're using a synchronized queue, then: no, you're safe! But you'll need to use the synchronized instance on both sides, the supplier and the feeder.
question 2: Terminating your worker thread when there is no work to do, is a simple job. However, you either way need a monitoring thread or have the queue start a background worker thread whenever the queue has something to do. The last one sounds more like the ActiveObject Pattern, than a simple queue (which's Single-Responsibily-Pattern says that it should only do queueing).
In addition, I'd go for a blocking queue instead of your code above. The way your code works requires CPU processing power even if there is no work to do. A blocking queue lets your worker thread sleep whenever there is nothing to do. You can have multiple sleeping threads running without using CPU processing power.
C# doesn't come with a blocking queue implementation, but there a many out there. See this example and this one.

Another option for making thread-safe use of queues is the ConcurrentQueue<T> class that has been introduced since 2009 (the year of this question). This may help avoid having to write your own synchronization code or at least help making it much simpler.
From .NET Framework 4.6 onward, ConcurrentQueue<T> also implements the interface IReadOnlyCollection<T>.