Parallel and Async Programming With C#

Rainer Stropek | software architects gmbh

Parallel & Async in C#

Agenda
Parallele und asynchrone Programmierung sind von einem Nischenthema
zu einer wichtigen Herausforderung für jeden .NET-Entwickler geworden.
Spätestens am Beispiel der Windows Runtime (WinRT), dem API für
Windows Store Apps, sieht man, welche Bedeutung dem Thema
beizumessen ist: In WinRT sind nahezu alle Funktionen, die etwas länger
dauern könnten, asynchron. Dadurch werden die vorhandenen
Ressourcen besser genutzt und das UI friert nie ein.

Rainer Stropek beginnt in seinem Workshop mit Grundlagen über parallele
und asynchrone Programmierung mit .NET. Er zeigt TPL und PLINQ in
Aktion. Darauf aufbauend geht er auf die neuen C#-Schlüsselwörter async
und await ein. Design-Guidelines für moderne, Task-basierende APIs sind
der Abschluss des Workshops. Sie erfahren, wie Task
Cancellation, Progress Reporting etc. richtig gemacht werden. Rainer stellt
zum Workshop ein Slidedeck zum Nachschlagen zur Verfügung, im
Workshop selbst stehen aber praktische Beispiele (C#) im Vordergrund.

Introduction
• software architects gmbh
• Rainer Stropek
– Developer, Speaker, Trainer
– MVP for Windows Azure
since 2010
– rainer@timecockpit.com
– @rstropek

http://www.timecockpit.com
http://www.timecockpit.com/devblog

Functional Programming
Concepts In C#
A Short Recap…

A short recap…
• Lambda expressions in C#
• Anonymous delegates
– Func, Func<T>
– Action, Action<T>
• Concepts of functional programming in C#
– Use first-class functions to reduce duplication of code
• Application of functional programming in LINQ
• BTW – This has been around since
C# 2 and/or 3.x

public delegate int MathOperation(int x, int y);

static void Main(string[] args)
{
MathOperation delegateOperation = Add;

// Anonymous delegate
MathOperation anonymousDelegateOperation = delegate(int num1, int num2)
{
return num1 + num2;
};

// Func<T>
Func<int, int, int> operationFunction = Add;

// Simple lambda functions
operationFunction = (x, y) => x + y;
delegateOperation = (x, y) => x + y;
}

public static int Add(int num1, int num2)
{
return num1 + num2;
}

public delegate string[] GenerateDataOperation(int numberOfElements);

static void Main(string[] args) {
GenerateDataOperation generateOperation = NameGenerator;

// Multi-line anonymous delegate
GenerateDataOperation anonymousGenerateOperation =
delegate(int numberOfElements) {
var result = new string[numberOfElements];
for (int i = 0; i < numberOfElements; i++) {
result[i] = string.Format("line_{0}", i);
}

return result;
};

// Multi-line lambda functions
Func<int, string[]> generatorFunction = (numberOfElements) => {
var result = new string[numberOfElements];
for (int i = 0; i < numberOfElements; i++) {
result[i] = string.Format("line_{0}", n);
}

return result;
};
}

Do we need some recap on LINQ?

• IEnumerable<T>
– Covariance
• C# LINQ syntax
• LINQ Support in .NET Framework
• Lambda expressions as data
– IQueryable<T>
– Expression trees

// Result of a method
var res5 = from training in demoData
select training.Attendees.First();

// Result of a property
var res6 = from training in demoData Map part of the famous
select training.Duration; map/reduce concept

// Anonymous type
select new { Even if you do not like var,
training.Title, here you have to use it
NumberOfAttendees =
training.Attendees.Count()
};

// Instance of existing type
select new TrainingInfo {
Title = training.Title,
NumberOfAttendees =
training.Attendees.Count()
};

Query vs. Method Syntax
• Linq queries are a concept of the language, not the .NET
Framework
• Compiler converts query into method calls

Pass first-class function as
parameter to a LINQ method

public static void SelectManyDemo()
{
var strings = new []
{
"This is string number one",
"This might be a second string" Method
}; Syntax

var result = strings
.SelectMany(s => s.Split(' ')) Reduce
.Distinct()
.Select(s => s.ToLower())
.OrderBy(s => s);
foreach (var line in result) Map
{
Console.WriteLine(line);
}
}

Method syntax
group training embedded in LINQ
by new { training.Title, query
NumberOfAttendees =
training.Attendees.Count(
a => a.CountryOfOrigin == "AT")
}
into trainingGroup
where trainingGroup.Key.NumberOfAttendees > 0
select new { trainingGroup.Key.Title,
AttendeesFromAustria =
trainingGroup.Key.NumberOfAttendees
};

Parallel Programming
Task Parallel Library

Goals
• Understand Tasks  foundation for
async/await
• Take a close look at C# 4.5‘s stars
async/await

Recommended Reading
• Joseph Albahari, Threading in C#
(from his O‘Reilly book C# 4.0 in a Nutshell)
• Patterns of Parallel Programming
• Task-based Asynchronous Pattern
• A technical introduction to the Async CTP
• Using Async for File Access
• Async Performance: Understanding the Costs
of Async and Await (MSDN Magazine)

See also Patterns of Parallel Programming

Multithreading

Pre .NET 4 .NET 4
• System.Threading • System.Threading.
Namespace Tasks Namespace
• Thread Klasse • Task und
Task<TResult>
Klassen
• ThreadPool Klasse • TaskFactory Klasse
• Parallel Klasse

Kurzer Überblick über Tasks
• Starten
– Parallel.Invoke(…) Nicht in Silverlight 
– Task.Factory.StartNew(…)
• Warten
– myTask.Wait()
– Task.WaitAll
– Task.WaitAny
– Task.Factory.ContinueWhenAll(…)
– Task.Factory.ContinueWhenAny(…)
• Verknüpfen
– Task.Factory.StartNew(…,
TaskCreationOptions.AttachedToParent);
• Abbrechen
– Cancellation Tokens

MSDN

Schleifen - Parallel.For
var source = new double[Program.Size];
var destination = new double[Program.Size];

Console.WriteLine(MeasuringTools.Measure(() => {
for (int i = 0; i < Program.Size; i++) {
source[i] = (double)i;
}

for (int i = 0; i < Program.Size; i++) {
destination[i] = Math.Pow(source[i], 2);
}
}));

Console.WriteLine(MeasuringTools.Measure(() => {
Parallel.For(0, Program.Size, (i) => source[i] = (double)i);
Parallel.For(0, Program.Size,
(i) => destination[i] = Math.Pow(source[i], 2));
}));

Schleifen - Parallel.For
• Unterstützung für Exception Handling
• Break und Stop Operationen
– Stop: Keine weiteren Iterationen
– Break: Keine Iterationen nach dem aktuellen Index mehr
– Siehe dazu auch ParallelLoopResult
• Int32 und Int64 Laufvariablen
• Konfigurationsmöglichkeiten (z.B. Anzahl an Threads)
• Schachtelbar
– Geteilte Threading-Ressourcen
• Effizientes Load Balancing
Nicht selbst entwickeln!
• U.v.m.

Schleifen - Parallel.ForEach
Console.WriteLine(
"Serieller Durchlauf mit foreach: {0}",
MeasuringTools.Measure(() =>
{
double sumOfSquares = 0;
foreach (var square in Enumerable.Range(0, Program.Size).Select(
i => Math.Pow(i, 2)))
{
sumOfSquares += square;
}
}));
Hoher Aufwand für
Console.WriteLine( abgesicherten Zugriff auf
"Paralleler Durchlauf mit foreach: {0}", MoveNext/Current
MeasuringTools.Measure(() =>  Parallele Version oft
{ langsamer
double sumOfSquares = 0;
Parallel.ForEach(Enumerable.Range(0, Program.Size)
.Select(i => Math.Pow(i, 2)), square => sumOfSquares += square);
}));

Von LINQ zu PLINQ
LINQ PLINQ
var result = source var result = source
.Where(…) .AsParallel()
.Select(…) .Where(…)
.Select(…)

Aus IEnumerable wird
ParallelQuery

Tipp: AsOrdered() erhält die
Sortierreihenfolge

Performancetipps für PLINQ
• Allokieren von Speicher in parallelem Lambdaausdruck vermeiden
– Sonst kann Speicher + GC zum Engpass werden
– Wenn am Server: Server GC
• False Sharing vermeiden
• Bei zu kurzen Delegates ist Koordinationsaufwand für Parallelisierung oft höher
als Performancegewinn
–  Expensive Delegates
– Generell: Auf richtige Granularität der Delegates achten
• AsParallel() kann an jeder Stelle im LINQ Query stehen
–  Teilweise serielle, teilweise parallele Ausführung möglich
• Über Environment.ProcessorCount kann Anzahl an Kernen ermittelt
werden
• Messen, Messen, Messen!

Was läuft hier falsch? (Code)
var result = new List<double>();
Console.WriteLine(
"Paralleler Durchlauf mit Parallel.ForEach: {0}",
{
Parallel.ForEach( Parallel.ForEach verwendet
source.AsParallel(), IEnumerable<T>  unnötige
i => Merge-Schritte
{
if (i % 2 == 0)
{
lock (result)
{
result.Add(i);
}
}
});
}));



Console.WriteLine(
"Paralleler Durchlauf mit Parallel.ForAll: {0}",
{
source.AsParallel().ForAll(
i =>
{
if (i % 2 == 0)
{
lock (result)
{
result.Add(i);
}
}
});
}));

Lock-free Collection wäre
überlegenswert!

Console.WriteLine(
"Serielles Lesen: {0}", Optimal für Parallelisierung
MeasuringTools.Measure(() => selbst bei einem Core (IO-Bound
{ Waits)
foreach (var url in urls)
{
var request = WebRequest.Create(url);
using (var response = request.GetResponse())
{
using (var stream = response.GetResponseStream())
{
var content = new byte[1024];
while (stream.Read(content, 0, 1024) != 0) ;
}
}
}
}));

Console.WriteLine( Anzahl Threads = Anzahl Cores;
"Paralleles Lesen: {0}",
könnte mehr sein, da IO-Bound
{ waits
Parallel.ForEach(urls, url =>
{
var request = WebRequest.Create(url);
using (var response = request.GetResponse())
{
using (var stream = response.GetResponseStream())
{
while (stream.Read(content, 0, 1024) != 0) ;
}
}
});
}));

Parallel.ForEach(
urls,
new ParallelOptions() { MaxDegreeOfParallelism = urls.Length },
url => { … });

Console.WriteLine(


"Paralleles Lesen: {0}",
{
urls.AsParallel().WithDegreeOfParallelism(urls.Length)
.Select(url => WebRequest.Create(url))
.Select(request => request.GetResponse())
.Select(response => new {
Response = response,
Stream = response.GetResponseStream() })
.ForAll(stream =>
{
while (stream.Stream.Read(content, 0, 1024) != 0) ;
stream.Stream.Dispose();
stream.Response.Close();
});
}));

OK für Client, tödlich für Server!
Wenn Anzahl gleichzeitiger User wichtig ist sind
andere Lösungen vorzuziehen.

private static void DoSomething()
{
Action<Action> measure = (body) =>
{
var startTime = DateTime.Now;
body();
Console.WriteLine("{0} {1}",
Thread.CurrentThread.ManagedThreadId, This process will run in
DateTime.Now - startTime); parallel
};

Action calcProcess = () =>
{ for (int i = 0; i < 100000000; i++);};

measure(() =>
Task.WaitAll(Enumerable.Range(0, 10)
.Select(i => Task.Run(() => measure(calcProcess)))
.ToArray()));
}
Note that we use the new Task.Run
function here; previously you had to use
Task.Factory.StartNew

Action<Action> measure = (body) => {
body();
Console.WriteLine("{0} {1}",
Thread.CurrentThread.ManagedThreadId,
DateTime.Now - startTime);
};

Action calcProcess = () =>
{ for (int i = 0; i < 350000000; i++);};
Action ioProcess = () =>
{ Thread.Sleep(1000); }; Note that this task is not
compute-bound
// ThreadPool.SetMinThreads(5, 5);
measure(() =>{
Task.WaitAll(Enumerable.Range(0, 10)
.Select(i => Task.Run(() => measure(ioProcess)))
.ToArray());
});

Action<Action> measure = (body) =>{
body();
Console.WriteLine("{0} {1}", Thread.CurrentThread.ManagedThreadId,
DateTime.Now - startTime);
};

Action calcProcess = () => { for (int i = 0; i < 350000000; i++);};
Action ioProcess = () => { Thread.Sleep(1000); };

ThreadPool.SetMinThreads(5, 5);
measure(() => Enumerable.Range(0, 10)
.AsParallel()
.WithDegreeOfParallelism(5)
.ForAll(i => measure(ioProcess)));

Excursus - PLINQ
• Use .AsParallel to execute LINQ
query in parallel
• Be careful if you care about ordering
– Use .AsOrdered if necessary
• Use .WithDegreeOfParallelism in
case of IO-bound tasks
• Use .WithCancellation to enable
cancelling

private static void DoSomethingElse()
{
Func<int, int> longRunningFunc = (prevResult) =>
{
Thread.Sleep(1000);
return prevResult + 42;
};

Concat tasks using ContinueWith

var task = Task.Run(() => longRunningFunc(0))
.ContinueWith(t => longRunningFunc(t.Result))
.ContinueWith(t => longRunningFunc(t.Result));
task.Wait();
Console.WriteLine(task.Result);
} Wait for completion of a
task.

Exception Handling
• AggregateException
• Remove nested Aggregate Exceptions with
Flatten
• Use CancellationToken for cooperative
cancellation
• Use the Handle method instead of loop over
Aggregate Exceptions
• Use Task.Exception

{
var child1 = Task.Factory.StartNew(() => {
var child2 = Task.Factory.StartNew(() => {
throw new MyCustomException("Attached child2 faulted.");
},

// Uncomment this line to see the exception rethrown.
// throw new MyCustomException("Attached child1 faulted.");
},
});

try {
task1.Wait();
}
catch (AggregateException ae) {

foreach (var e in ae.Flatten().InnerExceptions)
{
…
}
// or ...
// ae.Flatten().Handle((ex) => ex is MyCustomException);

}

var tokenSource = new CancellationTokenSource();
var token = tokenSource.Token;

{
CancellationToken ct = token;
while (someCondition)
{
// Do some work...
Thread.SpinWait(50000);
ct.ThrowIfCancellationRequested();
}
},
token);

// No waiting required.

foreach (var e in ae.InnerExceptions)
{
if (e is MyCustomException)
{
Console.WriteLine(e.Message);
}
else
{
throw;
}
}

ae.Handle((ex) =>
{
return ex is MyCustomException;
});

{
throw new MyCustomException("Task1 faulted.");
})
.ContinueWith((t) =>
{
Console.WriteLine("I have observed a {0}",
t.Exception.InnerException.GetType().Name);
},
TaskContinuationOptions.OnlyOnFaulted);

Thread Synchronisation
• Use C# lock statement to control access to
shared variables
– Under the hoods Monitor.Enter and
Monitor.Exit is used
– Quite fast, usually fast enough
– Only care for lock-free algorithms if really
necessary
• Note that a thread can lock the same object in
a nested fashion

// Source: C# 4.0 in a Nutshell, O‘Reilly Media
class ThreadSafe
{
static readonly object _locker = new object();
static int _val1, _val2;

static void Go()
{
lock (_locker)
{
if (_val2 != 0) Console.WriteLine (_val1 / _val2);
_val2 = 0;
}
}
}

// This is what happens behind the scenes
bool lockTaken = false;
try
{
Monitor.Enter(_locker, ref lockTaken);
// Do your stuff...
}
finally
{
if (lockTaken) Monitor.Exit(_locker);
}

// Provide a factory for instances of the Random class per thread
var tlr = new ThreadLocal<Random>(
() => new Random(Guid.NewGuid().GetHashCode()));

var watch = Stopwatch.StartNew();

var tasks =
// Run 10 tasks in parallel
Enumerable.Range(0, 10)
.Select(_ => Task.Run(() =>
// Create a lot of randoms between 0 and 9 and calculate
// the sum
Enumerable.Range(0, 1000000)
.Select(__ => tlr.Value.Next(10))
.Sum()))
.ToArray();
Task.WaitAll(tasks);

// Calculate the total
Console.WriteLine(tasks.Aggregate<Task<int>, int>(
0, (agg, val) => agg + val.Result));

Console.WriteLine(watch.Elapsed);
Do you think this is a good
watch = Stopwatch.StartNew(); solution?

// Provide a factory for instances of the Random class per thread
var tlr = new ThreadLocal<Random>(
() => new Random(Guid.NewGuid().GetHashCode()));

var watch = Stopwatch.StartNew();

Console.WriteLine(
ParallelEnumerable.Range(0, 10000000)
.Select(_ => tlr.Value.Next(10))
.Sum());

Console.WriteLine(watch.Elapsed);

Prefer PLINQ over TPL because it
automatically breaks the workload
into packages.

Alternatives For lock
• Mutex
• Semaphore(Slim)
• ReaderWriterLock(Slim)
• Not covered here in details

Thread Synchronization
• AutoResetEvent
– Unblocks a thread once when it receives a signal from another
thread
• ManualResetEvent(Slim)
– Like a door, opens and closes again
• CountdownEvent
– New in .NET 4
– Unblocks if a certain number of signals have been received
• Barrier class
– New in .NET 4
– Not covered here
• Wait and Pulse
– Not covered here

Producer/Consumer
var buffer = new Queue<long>();
var cancelTokenSource = new CancellationTokenSource();
var done = false;

var producer = Task.Factory.StartNew((cancelTokenObj) => {
var counter = 10000000;
var cancelToken = (CancellationToken)cancelTokenObj;
try {
while (!cancelToken.IsCancellationRequested && counter-- > 0) {
// Here we get some data (e.g. reading it from a file)
var value = DateTime.Now.Ticks;
// Write it to buffer with values that have to be processed
buffer.Enqueue(value);
}
}
finally {
done = true;
} buffer wird nicht gelockt
}, cancelTokenSource.Token);

Producer/Consumer

var consumer = Task.Factory.StartNew((cancelTokenObj) =>
{
while (!cancelToken.IsCancellationRequested && !done)
{
// Get the next value to process
lock (buffer)
{ Prüfung ob leer fehlt
var value = buffer.Dequeue();
}

// Here we do some expensive procesing
}
Consumer ist viel langsamer als
Producer  Producer
überschwemmt Consumer mit Daten

Collections für parallele
Programmierung
• System.Collections.Concurrent für Thread-
Safe Collections
– BlockingCollection<T>
Blocking und Bounding-Funktionen
– ConcurrentDictionary<T>
– ConcurrentQueue<T>
– ConcurrentStack<T>
– ConcurrentBag<T>
• Optimal zur Umsetzung von Pipelines
– Datei wird gelesen, gepackt, verschlüsselt, geschrieben

Producer/Consumer
var buffer = new BlockingCollection<long>(10);
var cancelTokenSource = new CancellationTokenSource();

var producer = Task.Factory.StartNew((cancelTokenObj) => {
var counter = 10000000;
try {
while (!cancelToken.IsCancellationRequested && counter-- > 0) {
// Here we get some data (e.g. reading it from a file)
var value = DateTime.Now.Ticks;
// Write it to the buffer with values that have to be processed
buffer.Add(value);
}


}
finally {
buffer.CompleteAdding();
}

Producer/Consumer

var consumer = Task.Factory.StartNew((cancelTokenObj) =>
{
foreach (var value in buffer.GetConsumingEnumerable())
{
if ( cancelToken.IsCancellationRequested )
{
break;
}


// Here we do some expensive procesing
}

Async Programming
The big new thing in C# 5

private static void DownloadSomeTextSync()
{
using (var client = new WebClient())
{
Console.WriteLine(
client.DownloadString(new Uri(string.Format(
"http://{0}",
(Dns.GetHostAddresses("www.basta.net"))[0]))));
}
}

Synchronous version of the code;
would block UI thread

private static void DownloadSomeText()
{
var finishedEvent = new AutoResetEvent(false);

// Notice the IAsyncResult-pattern here
Dns.BeginGetHostAddresses("www.basta.net", GetHostEntryFinished,
finishedEvent);
finishedEvent.WaitOne();
}

private static void GetHostEntryFinished(IAsyncResult result)
{
var hostEntry = Dns.EndGetHostAddresses(result);
{
// Notice the Event-based asynchronous pattern here
client.DownloadStringCompleted += (s, e) =>
{
Console.WriteLine(e.Result);
((AutoResetEvent)result.AsyncState).Set();
};
client.DownloadStringAsync(new Uri(string.Format(
"http://{0}",
hostEntry[0].ToString())));
} Notice that control flow is not clear
} any more.

private static void DownloadSomeText()
{
var finishedEvent = new AutoResetEvent(false);

// Notice the IAsyncResult-pattern here
Dns.BeginGetHostAddresses(
"www.basta.net",
(result) =>
{
var hostEntry = Dns.EndGetHostAddresses(result);
{
// Notice the Event-based asynchronous pattern here
client.DownloadStringCompleted += (s, e) =>
{
Console.WriteLine(e.Result);
((AutoResetEvent)result.AsyncState).Set();
};
client.DownloadStringAsync(new Uri(string.Format(
"http://{0}",
hostEntry[0].ToString())));
}
},
finishedEvent);
finishedEvent.WaitOne(); Notice how lambda expression
} can make control flow clearer

Notice the use of the new
Task Async Pattern APIs in
.NET 4.5 here
private static void DownloadSomeTextUsingTask()
{
Dns.GetHostAddressesAsync("www.basta.net")
.ContinueWith(t =>
{
{
return client.DownloadStringTaskAsync(new Uri(string.Format(
"http://{0}",
t.Result[0].ToString())));
}
})
.ContinueWith(t2 => Console.WriteLine(t2.Unwrap().Result))
.Wait();
}

Notice the use of lambda Notice how code has become
expressions all over the methods shorter and more readable

Rules For Async Method Signatures
• Method name ends with Async
• Return value
– Task if sync version has return type void
– Task<T> if sync version has return type T
• Avoid out and ref parameters
– Use e.g. Task<Tuple<T1, T2, …>> instead

// Synchronous version
private static void DownloadSomeTextSync()
{
{
Console.WriteLine(
client.DownloadString(new Uri(string.Format(
"http://{0}",
(Dns.GetHostAddresses("www.basta.net"))[0]))));
}
}
Notice how similar the sync and
async versions are!

// Asynchronous version
private static async void DownloadSomeTextUsingTaskAsync()
{
{
Console.WriteLine(
await client.DownloadStringTaskAsync(new Uri(string.Format(
"http://{0}",
(await Dns.GetHostAddressesAsync("www.basta.net"))[0]))));
}
}

private static async void DownloadSomeTextUsingTaskAsync2()
{
{
try
{
var ipAddress = await Dns.GetHostAddressesAsync("www.basta.net");
var content = await client.DownloadStringTaskAsync(
new Uri(string.Format("htt://{0}", ipAddress[0])));
Console.WriteLine(content);
}
catch (Exception)
{
Console.WriteLine("Exception!");
}
}
}

Let’s check the
generated code and
debug the async
code

Guidelines for async/await
• If Task ended in Canceled
state, OperationCanceledExceptio
n will be thrown

private async static void CancelTask()
{
try
{
var cancelSource = new CancellationTokenSource();
var result = await DoSomethingCancelledAsync(cancelSource.Token);
Console.WriteLine(result);
}
catch (OperationCanceledException)
{
Console.WriteLine("Cancelled!");
}
}

private static Task<int> DoSomethingCancelledAsync(CancellationToken token)
{
// For demo purposes we ignore token and always return a cancelled task
var result = new TaskCompletionSource<int>();
result.SetCanceled();
return result.Task;
}
Note usage of
TaskCompletionSource<T> here

Note that async API of WebClient uses
existing cancellation logic instead of
CancellationTokenSource

Parallel and Async Programming With C#

Guidelines for async/await
• Caller runs in parallel to awaited
methods
• Async methods sometimes do not run
async (e.g. if task is already completed
when async is reached)

Guidelines for async/await (UI Layer)

• async/await use SynchronizationContext to execute the
awaiting method  UI thread in case of UI layer
• Use Task.ConfigureAwait to disable this behavior
– E.g. inside library to enhance performance

public partial class MainWindow : Window
{
public MainWindow()
{
this.DataContext = this;
this.ListBoxContent = new ObservableCollection<string>();
this.InitializeComponent();
this.ListBoxContent.Add("Started");

this.Loaded += async (s, e) =>
{
for (int i = 0; i < 10; i++)
{
ListBoxContent.Add(await Task.Run(() =>
{
Thread.Sleep(1000);
return "Hello World!";
}));
}

this.ListBoxContent.Add("Finished");
};
}

public ObservableCollection<string> ListBoxContent { get; private set; }

Guidelines For Implementing
Methods Ready For async/await
• Return Task/Task<T>
• Use postfix Async
• If method support cancelling, add parameter of type
System.Threading.CancellationToken
• If method support progress reporting, add
IProgress<T> parameter
• Only perform very limited work before returning to the
caller (e.g. check arguments)
• Directly throw exception only in case of usage errors

public class Program : IProgress<int>
{
static void Main(string[] args)
{
var finished = new AutoResetEvent(false);
PerformCalculation(finished);
finished.WaitOne();
}

private static async void PerformCalculation(AutoResetEvent finished)
{
Console.WriteLine(await CalculateValueAsync(
42,
CancellationToken.None,
new Program()));
finished.Set();
}

public void Report(int value)
{
Console.WriteLine("Progress: {0}", value);
}

private static Task<int> CalculateValueAsync(
int startingValue,
CancellationToken cancellationToken,
IProgress<int> progress)
{
if (startingValue < 0)
{
// Usage error
throw new ArgumentOutOfRangeException("startingValue");
}

return Task.Run(() =>
{
int result = startingValue;
for (int outer = 0; outer < 10; outer++)
{
cancellationToken.ThrowIfCancellationRequested();

// Do some calculation
Thread.Sleep(500);
result += 42;

progress.Report(outer + 1);
}

return result;
}); Note that this pattern is good for
} compute-bound jobs
}

private static async void PerformCalculation(AutoResetEvent finished)
{
try
{
var cts = new CancellationTokenSource();
Task.Run(() =>
{
Thread.Sleep(3000);
cts.Cancel();
});
var result = await CalculateValueAsync(
42,
cts.Token,
new Program());
}
catch (OperationCanceledException)
{
Console.WriteLine("Cancelled!");
}

finished.Set();
}
Note cancellation and handling of
OperationCanceledException.

private static Task<int> CalculateValueAsync(
int startingValue,
CancellationToken cancellationToken,
IProgress<int> progress)
{
if (startingValue < 0)
{
// By definition the result has to be 0 if startingValue < 0
return Task.FromResult(0);
}

return Task.Run(() =>
{
[…]
});
}

Note that you could use Note how Task.FromResult is used
TaskCompletionSource instead to return a pseudo-task

Console.WriteLine(
"Paralleles Lesen mit TaskFactory: {0}",
{
var tasks = new Task[urls.Length];
for (int i = 0; i < urls.Length; i++)
{
tasks[i] = Task.Factory.StartNew(() => ReadUrl(urls[i]));
}

Task.WaitAll(tasks);
}
));
… Delegate verwendet Wert von i
aus dem Main Thread 
private static void ReadUrl(object url)
IndexOutOfRangeException
{
…
}

// Variante 1
… Durch lokale Variable wird delegate
var tasks = new Task[urls.Length]; unabhängig; mehr zum Thema unter dem
for (int i = 0; i < urls.Length; i++) Schlagwort Closures
{
var tmp = i;
tasks[i] = Task.Factory.StartNew(() => ReadUrl(urls[tmp]));
}
…

// Variante 2
var tasks = new Task[urls.Length]; State object wird an delegate übergeben
for (int i = 0; i < urls.Length; i++)
{
tasks[i] = Task.Factory.StartNew(ReadUrl, urls[i]);
}

for (var i = 0; i < 10; i++)
{
new Thread(() => Console.Write(i)).Start();
You have to be careful
}
with closures and
multi-threading.
for (var i = 0; i < 10; i++)
{
var temp = i;
new Thread(() => Console.Write(temp)).Start();
}

// Setup timer using an action (notice the closure here)
using (var timer = new System.Threading.Timer(
_ => progressIndicator(DateTime.Now - startTime), null, 1000, 1000))
{
Thread.Sleep(3100);
}

is the leading time tracking solution for knowledge
workers. Graphical time tracking calendar, automatic tracking of
your work using signal trackers, high level of extensibility and
customizability, full support to work offline, and SaaS
deployment model make it the optimal choice especially in the
IT consulting business.

Try for free and without any risk. You can get your
trial account at http://www.timecockpit.com. After the trial
period you can use for only 0,20€ per user and
month without a minimal subscription time and without a
minimal number of users.

ist die führende Projektzeiterfassung für
Knowledge Worker. Grafischer
Zeitbuchungskalender, automatische Tätigkeitsaufzeichnung
über Signal Tracker, umfassende Erweiterbarkeit und
Anpassbarkeit, volle Offlinefähigkeit und einfachste
Verwendung durch SaaS machen es zur Optimalen Lösung auch
speziell im IT-Umfeld.

Probieren Sie kostenlos und ohne Risko einfach
aus. Einen Testzugang erhalten Sie unter http://www.timecockpit.com.
Danach nutzen Sie um nur 0,20€ pro Benutzer und
Tag ohne Mindestdauer und ohne Mindestbenutzeranzahl.

Parallel and Async Programming With C#

More Related Content

What's hot (20)

Similar to Parallel and Async Programming With C# (20)

Recently uploaded (20)

Parallel and Async Programming With C#