Chapter 19
What can lead to a deadlock situation, and why is it undesirable?
A lack of (or mishandled) coordination between different lock objects can cause deadlock, in which no progress can be made and the program is locked in its current state.
How is the dining philosophers problem related to the problem of deadlock?
In the dining philosophers problem, as each philosopher is holding only one fork with their left hand, they cannot proceed to eat or put down the fork they are holding. The only way a philosopher gets to eat their food is for their neighbor philosopher to put their fork down, which is only possible if they can eat their own food; this creates a never-ending circle of conditions that can never be satisfied. This situation is, in essence, the nature of a deadlock, in which all elements of a system are stuck in place and no progress can be made.
What are the four Coffman conditions?
Deadlock is also defined by the necessary conditions that a concurrent program needs to have at the same time, in order for deadlock to occur. These conditions were first proposed by the computer scientist Edward G. Coffman, Jr., and are therefore known as the Coffman conditions. The conditions are as follows:
- At least one resource has to be in a non-shareable state. This means that that resource is being held by an individual process (or thread) and cannot be accessed by others; the resource can only be accessed and held by a single process (or thread) at any given time. This condition is also known as mutual exclusion.
- There exists one process (or thread) that is simultaneously accessing a resource and waiting for another held by other processes (or threads). In other words, this process (or thread) needs access to two resources in order to execute its instructions, one of which it is already holding, and the other of which it is waiting for from other processes (or threads). This condition is called hold and wait.
- Resources can only be released by a process (or a thread) holding them if there are specific instructions for the process (or thread) to do so. This is to say that unless the process (or thread) voluntarily and actively releases the resource, that resource remains in a non-shareable state. This is the no preemption condition.
- The final condition is called circular wait. As suggested by the name, this condition specifies that there exists a set of processes (or threads) such that the first process (or thread) in the set is in a waiting state for a resource to be released by the second process (or thread), which, in turn, needs to be waiting for the third process (or thread); finally, the last process (or thread) in the set is waiting for the first one.
How can resource ranking solve the problem of deadlock? What other problems occur when this is implemented?
Instead of accessing the resources arbitrarily, if the processes (or threads) are to access them in a predetermined, static order, the circular nature of the way that they acquire and wait for the resources will be eliminated. However, if you place enough locks on the resources of your concurrent program, it will become entirely sequential in its execution, and, combined with the overhead of concurrent programming functionalities, it will have an even worse speed than the purely sequential version of the program.
How can ignoring locks solve the problem of deadlock? What other problems can occur when this is implemented?
By ignoring locks, our program resources effectively become shareable among different processes/threads in a concurrent program, thus eliminating the first of the four Coffman conditions, mutual exclusion. Doing this, however, can be seen as misunderstanding the problem completely. We know that locks are utilized so that processes and threads can access the shared resources in a program in a systematic, coordinated way, to avoid mishandling the data. Removing any locking mechanisms in a concurrent program means that the likelihood of the shared resources, which are now free from accessing limitations, being manipulated in an uncoordinated way (and therefore becoming corrupted) increases significantly.
How is livelock related to deadlock?
In a livelock situation, the processes (or threads) in the concurrent program are able to switch their states, yet they simply switch back and forth infinitely, and no progress can be made.