This is the second article of two parts on dotNet threading. In this second part, I will discuss further the synchronization objects in the System.Threading dotNet namespace, thread local storage, COM interoperability and thread states.

This article is written for the intermediate and senior C# developer. Working knowledge of the C# programming language and dotNet framework is assumed. The article was written with a Beta version of VS.NET and associated documentation. Changes, although not anticipated, might occur before final release of VS.NET that invalidate portions of this article.

In the first article, I discussed how to create threads, thread pools and some of the synchronization objects available in the System.Threading dotNet namespace. In this second article, I will complete my discussion of the synchronization objects and will discuss thread local storage, COM interoperability and thread states.

Another popular design pattern introduced as a class in the dotNet framework is the ReaderWriterLock. This class allows an unlimited amount of read locks or one write lock, but not both. This allows anyone to read the protected resource, as long as nobody is writing to the protected resource and allows only one thread to write to the protected resource at any one time. Listing 1 presents a sample using the ReaderWriterLock class.

Listing 1: ReaderWriterLock Class

using System;

using System.Threading;

namespace ConsoleApplication9

{

   class Class1

   {

      public Class1()

      {

         rwlock = new ReaderWriterLock();

         val = "Writer Sequence Number is 1";

      }

      private ReaderWriterLock rwlock;

      private string val;

      public void Reader()

      {

         rwlock.AcquireReaderLock(Timeout.Infinite);

         Console.WriteLine("Acquired Read Handle: "

            "Value = {0}", val);

         Thread.Sleep(1);

         Console.WriteLine("Releasing Read Handle");

         rwlock.ReleaseReaderLock();

      }

      public void Writer()

      {

         rwlock.AcquireWriterLock(Timeout.Infinite);

         Console.WriteLine("Acquired Write Handle");

         int id = rwlock.WriterSeqNum;

         Console.WriteLine("Writer Sequence Number "

            "is {0}", id);

         Thread.Sleep(1);

         val = "Writer ";

         Thread.Sleep(1);

         val += "Sequence ";

         Thread.Sleep(1);

         val += "Number ";

         Thread.Sleep(1);

         val += "is ";

         Thread.Sleep(1);

         val += id;

         Console.WriteLine("Releasing Write Handle");

         rwlock.ReleaseWriterLock();

      }

      static void Main(string[] args)

      {

         Class1 obj = new Class1();

         const int n = 1000;

         Thread[] reader = new Thread[n];

         Thread[] writer = new Thread[10];

         for (int i=0;i<n;i++)

         {

            reader[i] = new Thread(

               new ThreadStart(obj.Reader));

            if (i<10)

            {

               writer[i] = new Thread(

                  new ThreadStart(obj.Writer));

            };

         }

         for (int i=0;i<n;i++)

         {

            reader[i].Start();

            if (i<10)

            {

               writer[i].Start();

            };

         }

      }

   }

}

In the above listing, I create 10 writer threads and 1000 reader threads. I parameterized the number of reader threads so that I could quickly trigger different behaviors in the code by modifying the number of reader threads. Once the threads are started they attempt to acquire read and write lock on the ReaderWriterLock object. If you run the code, then you can see the writer threads have a difficult time acquiring write locks. I tried to put as many small sleep statements as I could to force the threads to swap out of memory earlier than they would have normally.

The last synchronization object I’ll present here is the Mutex. The most useful feature of the Mutex class is that it may be named. This allows you to create two Mutex objects in different areas of code without having to share Mutex object instances. As long as the Mutex object instances have the same name, they will synchronize with each other. You could create the Mutex in two different processes on the same machine and the synchronization crosses the process boundary. Nor do you have to worry about passing the Mutex object in order to share the synchronization object between two threads or methods (see Listing 2).

Listing 2: Mutex Class

using System;

using System.Threading;

namespace ConsoleApplication10

{

   class Class1

   {

      public void ThreadStart()

      {

         Mutex mutex = new Mutex(false, "MyMutex");

         mutex.WaitOne();

         Console.WriteLine("Hello");

      }

      static void Main(string[] args)

      {

         Class1 obj = new Class1();

         Thread thread = new Thread(

            new ThreadStart(obj.ThreadStart));

         Mutex mutex = new Mutex(true, "MyMutex");

         thread.Start();

         Thread.Sleep(1000);

         Console.WriteLine("Signal");

         mutex.ReleaseMutex();

      }

   }

}

In the above listing, two separate Mutex objects are created, but the Mutex class allows the two instances to interact. The Signal will always precede the Hello in the output of this program. This is because the Mutex in the thread is created with the lock acquired. The second thread then creates the Mutex without acquiring the lock. The second thread will then wait on the mutex until the main thread releases the mutex a second later.

The Thread class and System.Threading namespace also contain some methods and classes for realizing thread local storage. Thread local storage is a manner of storing data in a container that is unique to the thread. Many threads could then use the same named container to store their data without concern of collision. Each thread’s local storage is distinct from another thread’s local storage and is only available in the one thread. Listing 3 shows a small sample using the thread-local-storage methods and classes.

Listing 3: Thread Local Storage

using System;

using System.Threading;

namespace ConsoleApplication11

{

   class Class1

   {

      public void ThreadStart()

      {

         string str1 = "My Cookie " +

            Thread.CurrentThread.GetHashCode();

         Console.WriteLine("worker thread: {0}",str1);

         LocalDataStoreSlot lds =

            Thread.GetNamedDataSlot("COOKIE");

         Thread.SetData(lds, str1);

         Thread.Sleep(1);

         LocalDataStoreSlot lds2 =

            Thread.GetNamedDataSlot("COOKIE");

         string str2 = "";

         str2 = (string)Thread.GetData(lds2);

         Console.WriteLine("worker thread: {0}",str2);

      }

      static void Main(string[] args)

      {

         string str1 = "My Cookie " +

            Thread.CurrentThread.GetHashCode();

         Console.WriteLine("main thread: {0}", str1);

         LocalDataStoreSlot lds =

            Thread.AllocateNamedDataSlot("COOKIE");

         Thread.SetData(lds, str1);

         Class1 obj = new Class1();

         Thread thread = new Thread(

            new ThreadStart(obj.ThreadStart));

         thread.Start();

         Thread.Sleep(1);

         LocalDataStoreSlot lds2 =

            Thread.GetNamedDataSlot("COOKIE");

         string str2 = "";

         str2 = (string)Thread.GetData(lds2);

         Console.WriteLine("main thread: {0}", str2);

      }

   }

}

You could also create and start more than one thread and the behavior of the thread local storage becomes more obvious. I have played with Win32 thread-local-storage functions and created my own for portability to UNIX, but I have rarely found them very useful. I strongly believe in stateless computing and thread-local-storage contradicts this belief.

Now what about those COM apartments? How do these new dotNet threads handle COM apartments? dotNet threads can reside in both single and multithreaded apartments. When a dotNet thread is first started it exists neither in a single-threaded or multithreaded apartment. A static state variable Thread.CurrentThread.Apartment indicates the current apartment type. If you run the code in Listing 4, then the apartment type will be Unknown, as the thread would not have entered an apartment yet.

Listing 4: Threading Model Attributes

using System;

using System.Threading;

namespace ConsoleApplication5

{

   class Class1

   {

//      line         output

//               // Unknown

//      [STAThread]      // STA

//      [MTAThread]      // MTA

      public static void Main(String[] args)

      {

         Console.WriteLine("Apartment State = {0}",

            Thread.CurrentThread.ApartmentState);

      }

   }

}

If you uncomment the line with the STAThread attribute, then the thread set its ApartmentState to STA. If you uncomment the line with the MTAThread attribute, then the thread set its ApartmentState to MTA. This allows control over the apartment type, similar to CoInitializeEx. You can also set the ApartmentState static member directly (see Listing 5).

Listing 5: ApartmentState

using System;

using System.Threading;

namespace ConsoleApplication6

{

   class Class1

   {

      static void Main(string[] args)

      {

//       Thread.CurrentThread.ApartmentState =

//          ApartmentState.STA;

         Thread.CurrentThread.ApartmentState =

            ApartmentState.MTA;

         Console.WriteLine("Apartment State = {0}",

            Thread.CurrentThread.ApartmentState);

      }

   }

}

Setting the ApartmentState property has the same affect as using the STAThread and MTAThread attributes.

There are also class attributes that affect the threading model used by the dotNet framework. The ThreadAffinity and Synchronization class attributes can be used to synchronize access to a class and its instance members.

[ThreadAffinity()]
public class Class1 : ContextBoundObject

[Synchronization()]
public class Class1 : ContextBoundObject

When calling into such classes, the calls are serialized to limit access to the class to one thread at any one time. At this point, these class context attributes are really thin on documentation. So, I’ll save a more in-depth explanation that may be incorrect anyway.

I figured with all this work I’m doing learning dotNet threads that I would leave you with an important resource. Table 1 shows my attempt in converting Win32 functions to dotNet classes and methods.

Table 1: Converting Win32 to dotNet

If you were to undertake a project of converting a Win32 application to a dotNet application, then this table could prove very useful. In some cases, a few objects and methods in the dotNet framework could closely emulate a Win32 function. I had to, on occasion, decide how closely they matched and sometimes decided that a match was not appropriate. As an example, you could create a semaphore with a Mutex object and a counter. But I wouldn’t say it’s a close match, so I didn’t mention these instances. In other cases, I had to decide between two matches.

The last few topics in this article are really just the few bits of reference information I dug up on dotNet threads. This section describes the states of a thread. The Thread object in the dotNet framework has a property called the ThreadState, which is one of the members of the following enumeration, which I pulled from the dotNET documentation.

public enum ThreadState
{
  Running = 0,
  SuspendRequested = 2,
  Background = 4,
  Unstarted = 8,
  WaitSleepJoin = 32,
  Suspended = 64,
  AbortRequested = 128,
  Aborted = 256
};

Unfortunately, I have been able to generate ThreadState’s that are not in this enumeration. Specifically, the Stopped ThreadState seems to be missing and is easy to generate. If you check the state of a thread that has run to completion, then the state is marked as Stopped.

What I also found is that it is quite easy to generate dual states. You can be in the AbortRequested state and the WaitSleepJoin state. If you catch the ThreadAbortException and then call Thread.Sleep, then the ThreadState will be “WaitSleepJoin, AbortRequested”, a dual state. The same is true if you are sleeping when the Suspend instance method is called. Immediately after the call to the Suspend instance method, the ThreadState property reports “SuspendRequested, WaitSleepJoin”, then quickly changes to “WaitSleepJoin, Suspended”.

I’ve encountered a few state diagrams that tried to depict the state transitions of dotNet threads. I must say that most are misleading or incomplete. The biggest problem is that most of the state diagrams did not attempt to account for dual states. My own attempt at the state diagram, I know, is still lacking but much further along then anything else I’ve seen (see Figure 1).

There is still a lot missing from the state diagram. Specifically, what happens when you Suspend(), Wait(), Join(), Sleep(), Abort() a background thread. I’m not going to confuse the diagram to explain these new states. Rather, let me explain that a thread is either a background thread or a foreground thread. Actions on a background thread are equivalent to actions on a foreground thread, except in one respect, which I will explain in the next paragraph. So, if you attempt to suspend a running background thread, then it will move to the SuspendRequested state, then to the Suspended state and finally back to the Background state, in the same manner as a foreground thread.

The difference between a background thread and a foreground thread is pretty simple. When the last foreground thread of a process is stopped, then the process terminates. There could be zero, 1 or an infinite number of background threads and they have no vote in whether a process terminates or not. So when the last foreground thread stops, then all background threads are also stopped and the process is stopped.

I’ve seen quite a few dot-NET programmers incorrectly use the background thread to mean any thread created using the Thread constructor. The terminology is therefore getting very confusing. The correct meaning of background thread in dotNet framework is a thread that does not have impact on whether a process is terminated.

Here’s a rather interesting tidbit of news. Many of the dotNet objects and types are thread-safe. The first time I heard that I was rather confused at what it could mean. Does this mean an increment (++) operation on a C# integer is atomic? I put together a small piece of C# code that launched a thousand threads and incremented and decremented one integer a million times per thread. I structured the code to swap the threads like mad to try and create a race condition that would invalidate the operations on the integer. I was unsuccessful in generating incorrect results. So, I assume the operation is atomic. But I don’t have any proof (beyond proof-by-example) that it is an atomic operation.

Throughout this article, I have written code that assumes that some operations on C# objects and types are atomic. I would never suggest writing such code in a production environment. In such an environment, you will have to fall back onto our old InterlockedIncrement and InterlockedDecrement friends. In C#, these are in the System.Threading.Interlocked class. The class has two static methods Interlocked.Increment and Interlocked.Decrement. Use them well.

I started this trek into dotNet threads for one reason. I wanted to evaluate them as a possible alternative for servers that require a lot of thread programming. What I found was that dotNet’s Threading namespace is by far the easiest way to write applications that require a lot of thread programming. I didn’t find any performance problems with the dotNet threads, but neither did I find them any faster than other thread libraries available in C++ or Java threads.

About the Author

Randy Charles Morin is the Lead Architect of SportMarkets Development from Toronto, Ontario, Canada and lives with his wife and two kids in Brampton, Ontario. He is the author of the www.kbcafe.com website, author of Wiley’s Programming Windows Services book and co-author of many other programming books and articles.