communication between threads

Hi.. I have created the threads for implementing different tasks... I want to exchange some data between them but i am really confused what should i use and what not
I can use pipes , ipc, shared memory, sockets but what is the best for my problem?

What kind of data do you want to exchange? (Small) messages, a big chunk of memory, or a stream of bytes? Is there the possibility that once your threads will be able to run on two different computers connected by a network?

I need to exchange small messages.. just to pass some stucts with data.... the threads will always run on the same pc..

Hello,

My suggestions:

1) If your threads run on the same machine, and in the same process address space, then I would suggest simply using global memory with access to the data controlled by semaphores or some such mechanism to insure that you won't have multiple writers touching the same memory at the same time. That is the easiest and induces the least overhead (which make it the fastest during run time).

2) If your threads run on the same machine, but run in different process address spaces, then I would suggest using shared memory with access to the data again controlled by semaphores or some such mechanism. This is also very easy to implement and induces just a little more overhead.

3) Lastly, if your threads will ever need to be run on separate machines, then I would suggest using TCP/IP sockets.

However, if you start out just needing to implement #1, but may eventually need to move to something like #3, then I would use sockets from the getgo. It really just boils down to how much overhead and time you want to spend in implementing the most cost effective algorithms for your specific requirements.

Hi,

I have a related question. I have an app that starts up several threads, and I want them to communicate via message queues. Currently, I am developing under Cygwin. The code compiles and links (with the '-Wall' option so it's strict), but I get a 'bad system call' runtime error as soon as the first call to msgget() is made:

// Create the projector inbox
printf("Creating projector inbox...");
projector_inbox_id = msgget(INBOX_PROJECTOR, 0666 | IPC_CREAT);
if (projector_inbox_id < 1) {
fprintf(stderr, "SmartHub: unable to create the projector inbox.\n");
exit(-1);
}


The code examples I see usually use the messaging API for inter-process communication, not inter-thread communication. Is that my problem?

Thanks,
Josh Fitzgerald

Can you try your program on a nativ linux machine? Or maybe you can try colinux.
Just to clarify: You run your programm under cygwin on a Windows machine, is that correct?
cygwin may not implement all system calls. You may check the cygwin manual.

Installation of colinux is a better way for using linux on windows, i guess.

Unfortunately we are a Windows shop and I cannot currently run this application natively (although I am working to remedy this). However, I have test apps that create threads and make use of mutexes and these ran without a hitch. This is what biased me towards the hypothesis that I did not understand how the message queue API was intended to be used.

You may have a look at colinux. It is called a cooperative os. it runs a complete unix kernel,
while you have your windows up and running. It is worth a look. However, installation is not
that easy i think. (i haven't tried by myself.)

you create some huge files (say 1 - 2GB) These will be used as virtual filesystems for the
linux box. In these files a linux system is installed and you run the colinux daemon and
there you go...

Be sure to design the app so that there is not a one-to-one correspondence between "user requests" and "threads to handle them." Instead, a pool of worker threads should be created, all of which sit and wait for a request, then do it, then check for another request or sleep until one arrives. This allows the total number of active threads to be predictable and constant, therefore manageable.

Wow, what a cool project! So I guess you are thinking my program should work, eh?