Home/ Questions/Q 286453
In Process

The Archive Base Latest Questions

Editorial Team
  • 0
Asked: 2026-05-12T05:38:16+00:00

In multitasking systems, some abnormal conditions prevent progress of executing processes or threads. I’ll

  • 0

In multitasking systems, some abnormal conditions prevent progress of executing processes or threads. I’ll refer to both processes and threads simply as "processes". Two of these conditions are called dead-lock and live-lock.

The former refers to processes which are blocking each other, thus preventing either from executing. The latter refers to processes which prevent each other from progressing, but do not actually block the execution. For instance, they might continually cause each other to rollback transactions, neither ever being able to finish them.

Another condition is known as resource starvation, in which one or more finite resources, required for the progress of the processes, have been depleted by them and can’t be restored unless the processes progress. This is also a special case of live-lock.

I’d like to know if there is any other definition, particularly an academic one, for "starvation" that is not limited to "resource starvation".

Share

Leave an answer

You must login to add an answer.

Need An Account,

1 Answer

  1. Editorial Team

    I wouldn’t say that resource starvation is a special case of a livelock. Usually:

    • In a livelock, no thread makes progress but they keep executing. (In a deadlock, they don’t even keep executing)

    • When starving, some thread(s) DO make progress and some thread(s) aren’t executing.

    A good explanation: http://docs.oracle.com/javase/tutorial/essential/concurrency/starvelive.html. But I understand the choice of terminology may vary.

    When it comes to starvation, the definition I heard is:

    Suppose it is possible to specify an infinite path of execution (interlace) consistent with assumptions (semaphore semantics, OS scheduler behaviour…) such that thread T is suspended waiting for some resource and never resumed, even if it was possible infinitely many times. Then T is called starving.

    But the practice doesn’t match that. Suppose two threads execute the same critical section in an infinite loop. Your synchronization code allows the first thread to enter the critical section once per hour. Is it starvation? Both threads are allowed to progress, but the first one is doing its work painfully slowly.

    The simplest source of starvation are weak semaphores. If you are using a synchronization primitive (or building your own) that behaves similarly, then starvation will result.

    Classical problems where starvation is well known:

    For more details, I wholeheartedly recommend The Little Book of Semaphores (free): http://www.greenteapress.com/semaphores/.

    You are asking if every starvation is caused by waiting for some resource. I’d say – yes.

    A thread can be suspended:

    (1) on some blocking system call – waiting on/acquiring a mutex, semaphore, conditional variable; write(), poll() etc.

    (2) on some nonblocking operation – like performing computations.

    Starving on (1) is starving on resources (mutexes, buffer etc.).

    Starving on (2) is starving on CPU – you can regard it as a resource. If it happens, the problem is with scheduler.

    HTH

    • 0