Monitor (synchronization)


In concurrent programming, a monitor is a synchronization construct that allows threads to have both mutual exclusion and the ability to wait for a certain condition to become false. Monitors also have a mechanism for signaling other threads that their condition has been met. A monitor consists of a mutex object and condition variables. A condition variable essentially is a container of threads that are waiting for a certain condition. Monitors provide a mechanism for threads to temporarily give up exclusive access in order to wait for some condition to be met, before regaining exclusive access and resuming their task.
Another definition of monitor is a thread-safe class, object, or module that wraps around a mutex in order to safely allow access to a method or variable by more than one thread. The defining characteristic of a monitor is that its methods are executed with mutual exclusion: At each point in time, at most one thread may be executing any of its methods. By using one or more condition variables it can also provide the ability for threads to wait on a certain condition. For the rest of this article, this sense of "monitor" will be referred to as a "thread-safe object/class/module".
Monitors were invented by Per Brinch Hansen and C. A. R. Hoare, and were first implemented in Brinch Hansen's Concurrent Pascal language.

Mutual exclusion

As a simple example, consider a thread-safe object for performing transactions on a bank account:
monitor class Account
While a thread is executing a method of a thread-safe object, it is said to occupy the object, by holding its mutex. Thread-safe objects are implemented to enforce that at each point in time, at most one thread may occupy the object. The lock, which is initially unlocked, is locked at the start of each public method, and is unlocked at each return from each public method.
Upon calling one of the methods, a thread must wait until no other thread is executing any of the thread-safe object's methods before starting execution of its method. Note that without this mutual exclusion, in the present example, two threads could cause money to be lost or gained for no reason. For example, two threads withdrawing 1000 from the account could both return true, while causing the balance to drop by only 1000, as follows: first, both threads fetch the current balance, find it greater than 1000, and subtract 1000 from it; then, both threads store the balance and return.
The syntactic sugar "monitor class" in the above example is implementing the following basic representation of the code, by wrapping each function's execution in mutexes:
class Account

Condition variables

Problem statement

For many applications, mutual exclusion is not enough. Threads attempting an operation may need to wait until some condition holds true. A busy waiting loop
while not do skip
will not work, as mutual exclusion will prevent any other thread from entering the monitor to make the condition true. Other "solutions" exist such as having a loop that unlocks the monitor, waits a certain amount of time, locks the monitor and check for the condition. Theoretically, it works and will not deadlock, but issues arise. It's hard to decide an appropriate amount of waiting time, too small and the thread will hog the CPU, too big and it will be apparently unresponsive. What is needed is a way to signal the thread when the condition P is true.

Case study: classic bounded producer/consumer problem

A classic concurrency problem is that of the bounded producer/consumer, in which there is a queue or ring buffer of tasks with a maximum size, with one or more threads being "producer" threads that add tasks to the queue, and one or more other threads being "consumer" threads that take tasks out of the queue. The queue is assumed to be non–thread-safe itself, and it can be empty, full, or between empty and full. Whenever the queue is full of tasks, then we need the producer threads to block until there is room from consumer threads dequeueing tasks. On the other hand, whenever the queue is empty, then we need the consumer threads to block until more tasks are available due to producer threads adding them.
As the queue is a concurrent object shared between threads, accesses to it must be made atomic, because the queue can be put into an inconsistent state during the course of the queue access that should never be exposed between threads. Thus, any code that accesses the queue constitutes a critical section that must be synchronized by mutual exclusion. If code and processor instructions in critical sections of code that access the queue could be interleaved by arbitrary context switches between threads on the same processor or by simultaneously-running threads on multiple processors, then there is a risk of exposing inconsistent state and causing race conditions.

Incorrect without synchronization

A naïve approach is to design the code with busy-waiting and no synchronization, making the code subject to race conditions:

global RingBuffer queue; // A thread-unsafe ring-buffer of tasks.
// Method representing each producer thread's behavior:
public method producer
// Method representing each consumer thread's behavior:
public method consumer

This code has a serious problem in that accesses to the queue can be interrupted and interleaved with other threads' accesses to the queue. The queue.enqueue and queue.dequeue methods likely have instructions to update the queue's member variables such as its size, beginning and ending positions, assignment and allocation of queue elements, etc. In addition, the queue.isEmpty and queue.isFull methods read this shared state as well. If producer/consumer threads are allowed to be interleaved during the calls to enqueue/dequeue, then inconsistent state of the queue can be exposed leading to race conditions. In addition, if one consumer makes the queue empty in-between another consumer's exiting the busy-wait and calling "dequeue", then the second consumer will attempt to dequeue from an empty queue leading to an error. Likewise, if a producer makes the queue full in-between another producer's exiting the busy-wait and calling "enqueue", then the second producer will attempt to add to a full queue leading to an error.

Spin-waiting

One naive approach to achieve synchronization, as alluded to above, is to use "spin-waiting", in which a mutex is used to protect the critical sections of code and busy-waiting is still used, with the lock being acquired and released in between each busy-wait check.

global RingBuffer queue; // A thread-unsafe ring-buffer of tasks.
global Lock queueLock; // A mutex for the ring-buffer of tasks.
// Method representing each producer thread's behavior:
public method producer
// Method representing each consumer thread's behavior:
public method consumer

This method assures that an inconsistent state does not occur, but wastes CPU resources due to the unnecessary busy-waiting. Even if the queue is empty and producer threads have nothing to add for a long time, consumer threads are always busy-waiting unnecessarily. Likewise, even if consumers are blocked for a long time on processing their current tasks and the queue is full, producers are always busy-waiting. This is a wasteful mechanism. What is needed is a way to make producer threads block until the queue is non-full, and a way to make consumer threads block until the queue is non-empty.

Condition variables

The solution is to use condition variables. Conceptually a condition variable is a queue of threads, associated with a monitor, on which a thread may wait for some condition to become true. Thus each condition variable is associated with an assertion. While a thread is waiting on a condition variable, that thread is not considered to occupy the monitor, and so other threads may enter the monitor to change the monitor's state. In most types of monitors, these other threads may signal the condition variable to indicate that assertion is true in the current state.
Thus there are three main operations on condition variables:
As a design rule, multiple condition variables can be associated with the same mutex, but not vice versa. This is because the predicate is the same for all threads using the monitor and must be protected with mutual exclusion from all other threads that might cause the condition to be changed or that might read it while the thread in question causes it to be changed, but there may be different threads that want to wait for a different condition on the same variable requiring the same mutex to be used. In the producer-consumer example, the queue must be protected by a unique mutex object, m. The "producer" threads will want to wait on a monitor using lock m and a condition variable which blocks until the queue is non-full. The "consumer" threads will want to wait on a different monitor using the same mutex m but a different condition variable which blocks until the queue is non-empty. It would never make sense to have different mutexes for the same condition variable, but this classic example shows why it often certainly makes sense to have multiple condition variables using the same mutex. A mutex used by one or more condition variables may also be shared with code that does not use condition variables, if those critical sections do not happen to require waiting for a certain condition on the concurrent data.

Monitor usage

The proper basic usage of a monitor is:

acquire; // Acquire this monitor's lock.
while
//... Critical section of code goes here...
signal; -- OR -- notifyAll; // cv2 might be the same as cv or different.
release; // Release this monitor's lock.

To be more precise, this is the same pseudocode but with more verbose comments to better explain what is going on:

//...
// About to enter the monitor.
// Acquire the advisory mutex associated with the concurrent
// data that is shared between threads,
// to ensure that no two threads can be preemptively interleaved or
// run simultaneously on different cores while executing in critical
// sections that read or write this same concurrent data. If another
// thread is holding this mutex, then this thread will be put to sleep
// and placed on m's sleep queue.
acquire;
// Now, we are holding the lock and can check the condition for the
// first time.
// The first time we execute the while loop condition after the above
// "acquire", we are asking, "Does the condition/predicate/assertion
// we are waiting for happen to already be true?"
while ) // "p" is any expression that checks the condition and
// evaluates to boolean. This itself is a critical
// section, so you *MUST* be holding the lock when
// executing this "while" loop condition!

// If this is not the first time the "while" condition is being checked,
// then we are asking the question, "Now that another thread using this
// monitor has notified me and woken me up and I have been context-switched
// back to, did the condition/predicate/assertion we are waiting on stay
// true between the time that I was woken up and the time that I re-acquired
// the lock inside the "wait" call in the last iteration of this loop, or
// did some other thread cause the condition to become false again in the
// meantime thus making this a spurious wakeup?
// The condition we are waiting for is true!
// We are still holding the lock, either from before entering the monitor or from
// the last execution of "wait".
// Critical section of code goes here, which has a precondition that our predicate
// must be true.
// This code might make cv's condition false, and/or make other condition variables'
// predicates true.
// Call signal/notify or notifyAll, depending on which condition variables'
// predicates have been made true or may have been made true,
// and the monitor semantic type being used.
for
// One or more threads have been woken up but will block as soon as they try
// to acquire m.
// Release the mutex so that notified thread and others can enter their critical
// sections.
release;

Solving the bounded producer/consumer problem

Having introduced the usage of condition variables, let's use it to revisit and solve the classic bounded producer/consumer problem. The classic solution is to use two monitors, comprising two condition variables sharing one lock on the queue:

global volatile RingBuffer queue; // A thread-unsafe ring-buffer of tasks.
global Lock queueLock; // A mutex for the ring-buffer of tasks.
global CV queueEmptyCV; // A condition variable for consumer threads waiting for the queue to
// become non-empty.
// Its associated lock is "queueLock".
global CV queueFullCV; // A condition variable for producer threads waiting for the queue
// to become non-full. Its associated lock is also "queueLock".
// Method representing each producer thread's behavior:
public method producer
// Method representing each consumer thread's behavior:
public method consumer

This ensures concurrency between the producer and consumer threads sharing the task queue, and blocks the threads that have nothing to do rather than busy-waiting as shown in the aforementioned approach using spin-locks.
A variant of this solution could use a single condition variable for both producers and consumers, perhaps named "queueFullOrEmptyCV" or "queueSizeChangedCV". In this case, more than one condition is associated with the condition variable, such that the condition variable represents a weaker condition than the conditions being checked by individual threads. The condition variable represents threads that are waiting for the queue to be non-full and ones waiting for it to be non-empty. However, doing this would require using notifyAll in all the threads using the condition variable and cannot use a regular signal. This is because the regular signal might wake up a thread of the wrong type whose condition has not yet been met, and that thread would go back to sleep without a thread of the correct type getting signalled. For example, a producer might make the queue full and wake up another producer instead of a consumer, and the woken producer would go back to sleep. In the complementary case, a consumer might make the queue empty and wake up another consumer instead of a producer, and the consumer would go back to sleep. Using notifyAll ensures that some thread of the right type will proceed as expected by the problem statement.
Here is the variant using only one condition variable and notifyAll:

global volatile RingBuffer queue; // A thread-unsafe ring-buffer of tasks.
global Lock queueLock; // A mutex for the ring-buffer of tasks.
global CV queueFullOrEmptyCV; // A single condition variable for when the queue is not ready for any thread
// -- i.e., for producer threads waiting for the queue to become non-full
// and consumer threads waiting for the queue to become non-empty.
// Its associated lock is "queueLock".
// Not safe to use regular "signal" because it is associated with
// multiple predicate conditions.
// Method representing each producer thread's behavior:
public method producer
// Method representing each consumer thread's behavior:
public method consumer

Synchronization primitives

Implementing mutexes and condition variables requires some kind of synchronization primitive provided by hardware support that provides atomicity. Locks and condition variables are higher-level abstractions over these synchronization primitives. On a uniprocessor, disabling and enabling interrupts is a way to implement monitors by preventing context switches during the critical sections of the locks and condition variables, but this is not enough on a multiprocessor. On a multiprocessor, usually special atomic read-modify-write instructions on the memory such as test-and-set, compare-and-swap, etc. are used, depending on what the ISA provides. These usually require deferring to spin-locking for the internal lock state itself, but this locking is very brief. Depending on the implementation, the atomic read-modify-write instructions may lock the bus from other cores' accesses and/or prevent re-ordering of instructions in the CPU. Here is an example pseudocode implementation of parts of a threading system and mutexes and Mesa-style condition variables, using test-and-set and a first-come, first-served policy. This glosses over most of how a threading system works, but shows the parts relevant to mutexes and condition variables:

Sample Mesa-monitor implementation with Test-and-Set


// Basic parts of threading system:
// Assume "ThreadQueue" supports random access.
public volatile ThreadQueue readyQueue; // Thread-unsafe queue of ready threads. Elements are.
public volatile global Thread* currentThread; // Assume this variable is per-core.
// Implements a spin-lock on just the synchronized state of the threading system itself.
// This is used with test-and-set as the synchronization primitive.
public volatile global bool threadingSystemBusy=false;
// Context-switch interrupt service routine :
// On the current CPU core, preemptively switch to another thread.
public method contextSwitchISR
// Thread sleep method:
// On current CPU core, a synchronous context switch to another thread without putting
// the current thread on the ready queue.
// Must be holding "threadingSystemBusy" and disabled interrupts so that this method
// doesn't get interrupted by the thread-switching timer which would call contextSwitchISR.
// After returning from this method, must clear "threadingSystemBusy".
public method threadSleep
public method wait
public method signal
public method broadcast
class Mutex
struct ConditionVariable

Semaphore

As an example, consider a thread-safe class that implements a semaphore.
There are methods to increment and to decrement a private integer s.
However, the integer must never be decremented below 0; thus a thread that tries to decrement must wait until the integer is positive.
We use a condition variable sIsPositive with an associated assertion of
monitor class Semaphore
Implemented showing all synchronization :
class Semaphore

Monitor implemented using semaphores

Conversely, locks and condition variables can also be derived from semaphores, thus making monitors and semaphores reducible to one another:
The implementation given here is incorrect. If a thread calls wait after broadcast has been called, an originally thread may be stuck indefinitely, since broadcast increments the semaphore only enough times for threads already waiting.

public method wait
public method signal
public method broadcast
class Mutex
class ConditionVariable

When a signal happens on a condition variable that at least one other thread is waiting on,
there are at least two threads that could then occupy the monitor:
the thread that signals and any one of the threads that is waiting. In order that at most
one thread occupies the monitor at each time, a choice must be made. Two schools of thought exist on how best to
resolve this choice. This leads to two kinds of condition variables which will be examined next:
The original proposals by C. A. R. Hoare and Per Brinch Hansen were for blocking condition variables. With a blocking condition variable, the signaling thread must wait outside the monitor until the signaled thread relinquishes occupancy of the monitor by either returning or by again waiting on a condition variable. Monitors using blocking condition variables are often called Hoare-style monitors or signal-and-urgent-wait monitors.
We assume there are two queues of threads associated with each monitor object
In addition we assume that for each condition variable, there is a queue
All queues are typically guaranteed to be fair and, in some implementations, may be guaranteed to be first in first out.
The implementation of each operation is as follows.
enter the monitor:
enter the method
if the monitor is locked
add this thread to e
block this thread
else
lock the monitor
leave the monitor:
schedule
return from the method
wait :
add this thread to.q
schedule
block this thread
signal :
if there is a thread waiting on.q
select and remove one such thread t from.q

add this thread to s
restart t

block this thread
schedule :
if there is a thread on s
select and remove one thread from s and restart it

else if there is a thread on e
select and remove one thread from e and restart it

else
unlock the monitor

The schedule routine selects the next thread to occupy the monitor
or, in the absence of any candidate threads, unlocks the monitor.
The resulting signaling discipline is known a "signal and urgent wait," as the signaler must wait, but is given priority over threads on the entrance queue. An alternative is "signal and wait," in which there is no s queue and signaler waits on the e queue instead.
Some implementations provide a signal and return operation that combines signaling with returning from a procedure.
signal and return :
if there is a thread waiting on.q
select and remove one such thread t from.q

restart t

else
schedule
return from the method
In either case, when a condition variable is signaled and there is at least one thread on waiting on the condition variable, the signaling thread hands occupancy over to the signaled thread seamlessly, so that no other thread can gain occupancy in between. If is true at the start of each signal operation, it will be true at the end of each wait operation. This is summarized by the following contracts. In these contracts, is the monitor's invariant.
enter the monitor:
postcondition
leave the monitor:
precondition
wait :
precondition
modifies the state of the monitor
postcondition and
signal :
precondition and
modifies the state of the monitor
postcondition
signal and return :
precondition and
In these contracts, it is assumed that and do not depend on the
contents or lengths of any queues.
or
modifies the state of the monitor
postcondition
See Howard and Buhr et al., for more).
It is important to note here that the assertion is entirely up to the programmer; he or she simply needs to be consistent about what it is.
We conclude this section with an example of a thread-safe class using a blocking monitor that implements a bounded, thread-safe stack.
monitor class SharedStack
Note that, in this example, the thread-safe stack is internally providing a mutex, which, as in the earlier producer/consumer example, is shared by both condition variables, which are checking different conditions on the same concurrent data. The only difference is that the producer/consumer example assumed a regular non-thread-safe queue and was using a standalone mutex and condition variables, without these details of the monitor abstracted away as is the case here. In this example, when the "wait" operation is called, it must somehow be supplied with the thread-safe stack's mutex, such as if the "wait" operation is an integrated part of the "monitor class". Aside from this kind of abstracted functionality, when a "raw" monitor is used, it will always have to include a mutex and a condition variable, with a unique mutex for each condition variable.

Nonblocking condition variables

With nonblocking condition variables, signaling does not cause the signaling thread to lose occupancy of the monitor. Instead the signaled threads are moved to the e queue. There is no need for the s queue.
With nonblocking condition variables, the signal operation is often called notify — a terminology we will follow here. It is also common to provide a notify all operation that moves all threads waiting on a condition variable to the e queue.
The meaning of various operations are given here.
enter the monitor:
enter the method
if the monitor is locked
add this thread to e
block this thread
else
lock the monitor
leave the monitor:
schedule
return from the method
wait :
add this thread to.q
schedule
block this thread
notify :
if there is a thread waiting on.q
select and remove one thread t from.q

move t to e
notify all :
move all threads waiting on.q to e
schedule :
if there is a thread on e
select and remove one thread from e and restart it
else
unlock the monitor
As a variation on this scheme, the notified thread may be moved to a queue called w, which has priority over e. See Howard and Buhr et al. for further discussion.
It is possible to associate an assertion with each condition variable such that is sure to be true upon return from wait . However, one must
ensure that is preserved from the time the notifying thread gives up occupancy until the notified thread is selected to re-enter the monitor. Between these times there could be activity by other occupants. Thus it is common for to simply be true.
For this reason, it is usually necessary to enclose each wait operation in a loop like this
while not do wait c
where is some condition stronger than. The operations notify and notify all are treated as "hints" that may be true for some waiting thread.
Every iteration of such a loop past the first represents a lost notification; thus with nonblocking monitors, one must be careful to ensure that too many notifications can not be lost.
As an example of "hinting" consider a bank account in which a withdrawing thread will wait until the account has sufficient funds before proceeding
monitor class Account
In this example, the condition being waited for is a function of the amount to be withdrawn, so it is impossible for a depositing thread to know that it made such a condition true. It makes sense in this case to allow each waiting thread into the monitor to check if its assertion is true.

Implicit condition variable monitors

In the Java language, each object may be used as a monitor. Methods requiring mutual exclusion must be explicitly marked with the synchronized keyword. Blocks of code may also be marked by synchronized.
Rather than having explicit condition variables, each monitor is equipped with a single wait queue in addition to its entrance queue. All waiting is done on this single wait queue and all notify and notifyAll operations apply to this queue. This approach has been adopted in other languages, for example C#.

Implicit signaling

Another approach to signaling is to omit the signal operation. Whenever a thread leaves the monitor the assertions of all waiting threads are evaluated until one is found to be true. In such a system, condition variables are not needed, but the assertions must be explicitly coded. The contract for wait is
wait :
precondition
modifies the state of the monitor
postcondition and

History

Brinch Hansen and Hoare developed the monitor concept in the early 1970s, based on earlier ideas of their own and of Edsger Dijkstra. Brinch Hansen published the first monitor notation, adopting the class concept of Simula 67, and invented a queueing mechanism. Hoare refined the rules of process resumption. Brinch Hansen created the first implementation of monitors, in Concurrent Pascal. Hoare demonstrated their equivalence to semaphores.
Monitors were soon used to structure process synchronization in the Solo operating system.
Programming languages that have supported monitors include
A number of libraries have been written that allow monitors to be constructed in languages that do not support them natively. When library calls are used, it is up to the programmer to explicitly mark the start and end of code executed with mutual exclusion. Pthreads is one such library.