Passing data structures to different threads

Question

I have an application that will be spawning multiple threads. However, I feel there might be an issue with threads accessing data that they shouldn't be. I'm relatively new to threading so bare with me.

Here is the structure of the threaded application (sorry for the crudeness):

                   MainThread
                  /          \
                 /            \
                /              \
        Thread A               Thread B
       /        \              /       \
      /          \            /         \
     /            \          /           \
Thread A_1     Thread A_2   Thread B_1    Thread B_2

Under each lettered thread (which could be many), there will only be two threads and they are fired of sequentially. The issue i'm having is I'm not entirely sure how to pass in a datastructure into these threads.

So, the datastructure is created in MainThread, will be modified in the lettered thread (Thread A, etc) specific to that thread and then a member variable from that datastructure is sent to Letter_Numbered threads.

Currently, the lettered thread class has a member variable and when the class is constructed, the datastructure from mainthread is passed in by reference, invoking the copy constructor so the lettered thread has it's own copy to play with.

The lettered_numbered thread simply takes in a string variable from the data structure within the lettered thread. My question is, is this accceptable? Is there a much better way to ensure each lettered thread gets its own data structure to play with?

Sorry for the somewhat poor explanation, please leave comments and i'll try to clarify.

EDIT: So my lettered thread constructor should take the VALUE of the data structure, not the reference?

Answer 1

I would have each thread create it's own copy of the datastructure, e.g. you pass the structure in the constructor and then explicitly create a local copy. Then you are guaranteed that the threads have distinct copies. (You say that it's passsed by reference, and that this invokes the copy constructor. I think you mean pass by value? I feel it's better to explicitly make a copy, to leave no doubt and to make your intent clear. Otherwise someone might later come along and change your pass by value to pass by reference as a "smart optimization".)

EDIT: Removed comment about strings. For some reason, I was assuming .NET.

To ensure strings are privately owned, follow the same procedure, create a copy of the string, which you can then freely modify.

Answer 2

Have you looked at boost threads?

You would basically create a callable class that has a constructor that takes the parameters the thread is to work on and then launch the thread by passing objects of your callable class, initialized and ready to go.

This is very similar to how Java implements threads and it makes a good amount of sense most of the time from a design point of view.

Answer 3

There is a pattern called Active Object Pattern wherein each object executes in its own thread. Frameworks like ACE support this. If you have access to such frameworks, you should use those. In any case, i would believe creating a new instance of an object and allowing it to exetute in its own thread is much cleaner that invoking the copy-constructor to make a copy of the object. Else see if you can fit a solution that uses Thread Local Storage.

Answer 4

You aparently are making a copy of the data for each trhead and everything works? then no problem.

Here are some additional thoughts:

• If data is read only, you can share a single struct and everything will be ok, as long as each read is small and fast (basic types)
• If data needs to be written, but "private" (or contained) to each thread, then send a copy to each thread (what you are doing). Caveat: I assume the data is not too big and a copy does not eat to much resources.
• If the data needs to be written and the new values shared between threads, then you need to think about it (read on it) and create a proper design. I like a transactional object to centralize each threads read/write operation. Like a tiny database in memory. Check on thread mutex, semaphores and critical sections). Dealing with huge data set I have used a database to centralize requests (See ODBM). You can also check existing messaging queuing libraries (like MSMQ) to have data change ordered and synchronized.

Hope this helps.

Answer 5

It seems unlikely that you would want each thread to operate on the data and then not at least occasionally have another thread react to what another thread has done to another thread's work on the data. If you are truly independent meaning that no other thread truly will ever care about work that another thread has done, then I suggest making a copy of the data, otherwise in the case where you will want to do work in one thread and make that result of that work available to another thread I would suggest that you, pass a reference/pointer to the object around and then protect access to it via locks so that the threads can work with it, properly, I suggest a multi-read, single writer lock implementation.