• 0

Using semaphores


Question

Hello, this is the first time I need to use semaphores, yes it is for an assignment. I have trouble wrapping my head around the concept. Come to think of it, wrapping a solid spheroidal body around an idea isn't trivial by itself, but anyway I disgress.

I have to write a little program that simulates a network. There are several computers in the network. Only one computer can be executing at any given time, and they must execute one after the other. Also, each computer is responsible for telling the "next" computer that it is its time to execute.

I am supposed to use semaphores to synchronize this. Now I don't see really how can semaphores help me do this. My first idea would be to make a "cycle" object, protected with a mutex, so only one computer can own the cycle at any given time. The computer could write in a global variable the id of the next computer and unlock the mutex when it's done. However, we're supposed to use semaphores, not mutexes.

Oh and it's POSIX semaphores and pthreads btw.

Thanks.

Link to comment
https://www.neowin.net/forum/topic/749492-using-semaphores/
Share on other sites

13 answers to this question

Recommended Posts

  • 0

I only covered semaphores on my course a few weeks back but I think the basic principle is the same as the mutex in that it creates a critical region that only one thread/process can access at one.

For example:

You spawn 2 threads to carry out some job, but only one can execute at the one time. Initially you set the semaphore to 1.

The first thread to call wait(sem) decrements the semaphore and is allowed to continue. The slower of the two threads is blocked at the wait statement because sem cannot be < 0. When the first thread is finished the work it is to carry out, it performs a signal(sem) which increments the semaphore, allowing the second thread to decrement it, and continue. This could also be looped so they can pass it back and forth between them. You will probably need a small wait after the signal though however to ensure that the other thread has enough time to perform the wait.

Think that's right. Hope it helps.

  • 0

Hmm, so you know how to do this using a mutex but not a semaphore? A mutex is basically just a binary semaphore. Of course since you said you need to control the order, right? As in there's a specific ordering to the "computers" and you want to loop through each, activating them in sequence? So couldn't you just have each node in the network provide its own semaphore that it blocks on until the preceding node signals it?

Or are you supposed to using a counting semaphore somehow, or only signal computers that have requested to be called when its their turn?

  • 0
  Quote
Hmm, so you know how to do this using a mutex but not a semaphore? A mutex is basically just a binary semaphore.
Well I know how pthread_mutex works but I never used semaphore.h. Also, I understand the concept of a mutex (locking/unlocking a resource), but I don't get the concept of a "non-binary" one. What, so you can lock the resource a certain number of times before it's, uh, really locked? :unsure:
  Brandon Live said:
Of course since you said you need to control the order, right? As in there's a specific ordering to the "computers" and you want to loop through each, activating them in sequence?
Yes exactly. It should go like this :

Computer 1 does cycle #1

Computer 2 does cycle #1

Computer 3 does cycle #1

Main thread does some checking routine

Computer 1 does cycle #2

Copmuter 2 does cycle #2

etc.

Each computer is its own thread.

  Quote
So couldn't you just have each node in the network provide its own semaphore that it blocks on until the preceding node signals it?
Can you explain your idea a bit further please ?
  Quote
Or are you supposed to using a counting semaphore somehow, or only signal computers that have requested to be called when its their turn?
It just says "use semaphors to order the computers". Edited by Dr_Asik
  • 0
  Dr_Asik said:
  Quote
So couldn't you just have each node in the network provide its own semaphore that it blocks on until the preceding node signals it?

Can you explain your idea a bit further please ?

I think he means that the first task signals the semaphore for task 2 so it can begin.

Maybe something like this?

Thread 1

task1 = 1;

wait(task1);

signal(task2);

Thread 2

The task2 semapohre is initially set to 0 until it is signalled from thread 1, so thread 2 blocks at the wait statement.

wait(task2);

.

.

.

.

signal(task3);

etc.

  • 0

Relinquishing the semaphore key is not a good idea, whos to say computer 3 wont get the key when its computer 2 that wants it.

C subroutines to checkout

sem_post(), sem_wait()

signals.h

in this u can do something like kill(PID, SIGINT); which will send the pid(computer#) a signal interrupt.

..
	signal(SIGINT, s_handler);
...

void s_handler(int sig)
{
	   switch(sig)
	   {
		case SIGINT:
			 //grab key and begin execution
		 }
}

once u set up the signals to recieve sigint, you can communicate with processes to grab the semaphore key instead of all processes waiting for it.

This topic is now closed to further replies.
  • Recently Browsing   0 members

    • No registered users viewing this page.
  • Posts

    • Since the .88 drivers were released yesterday, I don't believe that these benchmarks use the new drivers at all. Also note that these benchmarks are clearly just non-DLSS with Frame Gen tests, looking only at raw throughput of the core GPU die. That's fair, but should be identified as such. And, the whole point of the 50xx series is DLSS4 with Multi-Frame Gen which will 2x-4x frame rates, which means extra headroom is used for higher quality/resolution settings. Ignoring all of this, just so one can compare a new technology product to the low to mid-range competition, is disingenuous to say the least. And repeating it in this "we'll do anything to pretend the low to mid-range products are actually competitive" article without acknowledging these obvious handicaps and caveats is a disservice. Shame on you.
    • Its a couple things really. Windows 11 had a VERY rocky start. 21H2 was basically beta-level garbage, and 22H2 only slightly improved on that. A lot of the anti-11 people based their options on those early builds. In adiation to that, Windows 11 does have a really dumb start menu that deserves hate, but I do agree that many of the comments about the entire UI being unusable are exaugurated and shortsighted. Last, and this is big for gaming, Windows 11 in its default state, simply doesn't perform as well as Windows 10 does. It takes a lot of customization, and disabling some of that security you mentioned, to get Windows 11 to the same level as Windows 10. There are a few outliers where Windows 11 does excel, storage I/O being one, but that doesn't matter for most people.
    • msg * "Connect your VPN....and also learn Powershell"  
    • Reminder: Because this survey is voluntary and doesn't track specific user changes over time, it is worthless for comparisons month to month or year to year. Therefore it is nothing but pure monthly clickbait garbage. Thanks to pressure from users, Neowin used to have a disclaimer about this, but doesn't anymore. Valve could easily release the actual aggregated information (anonymized for privacy) , but I assume that is only for advertisers/etc. that pay for it...
  • Recent Achievements

    • Week One Done
      Devesh Beri earned a badge
      Week One Done
    • Week One Done
      956400 earned a badge
      Week One Done
    • First Post
      loose_observer earned a badge
      First Post
    • Week One Done
      BeeJay_Balu earned a badge
      Week One Done
    • Week One Done
      filminutz earned a badge
      Week One Done
  • Popular Contributors

    1. 1
      +primortal
      450
    2. 2
      ATLien_0
      158
    3. 3
      +FloatingFatMan
      150
    4. 4
      Nick H.
      65
    5. 5
      +thexfile
      62
  • Tell a friend

    Love Neowin? Tell a friend!