Grand Central Dispatch works on the concept of "Dispatch Queues". A dispatch queue executes tasks you designate in the order which they are passed. There are three types of dispatch queues:

• Serial Dispatch Queues (aka private dispatch queues) execute one task at a time, in order. They are frequently used to synchronize access to a resource.
• Concurrent Dispatch Queues (aka global dispatch queues) execute one or more tasks concurrently.
• The Main Dispatch Queue executes tasks on the main thread.

To access the main queue:

let mainQueue = DispatchQueue.main

let mainQueue = dispatch_get_main_queue()

The system provides concurrent global dispatch queues (global to your application), with varying priorities. You can access these queues using the DispatchQueue class in Swift 3:

let globalConcurrentQueue = DispatchQueue.global(qos: .default)

equivalent to

let globalConcurrentQueue = DispatchQueue.global()

let globalConcurrentQueue = dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT, 0)

In iOS 8 or later, the possible quality of service values which may be passed are .userInteractive, .userInitiated, .default, .utility, and .background. These replace the DISPATCH_QUEUE_PRIORITY_ constants.

You can also create your own queues with varying priorities:

let myConcurrentQueue = DispatchQueue(label: "my-concurrent-queue", qos: .userInitiated, attributes: [.concurrent], autoreleaseFrequency: .workItem, target: nil)
let mySerialQueue = DispatchQueue(label: "my-serial-queue", qos: .background, attributes: [], autoreleaseFrequency: .workItem, target: nil)

let myConcurrentQueue = dispatch_queue_create("my-concurrent-queue", DISPATCH_QUEUE_CONCURRENT)
let mySerialQueue = dispatch_queue_create("my-serial-queue", DISPATCH_QUEUE_SERIAL)

In Swift 3, queues created with this initializer are serial by default, and passing .workItem for autorelease frequency ensures an autorelease pool is created and drained for each work item. There is also .never, which means you will be managing your own autorelease pools yourself, or .inherit which inherits the setting from the environment. In most cases you probably won't use .never except in cases of extreme customization.

GCD provides mechanism for performing a loop, whereby the loops happen concurrently with respect to each other. This is very useful when performing a series of computationally expensive calculations.

Consider this loop:

for index in 0 ..< iterations {
    // Do something computationally expensive here
}

You can perform those calculations concurrently using concurrentPerform (in Swift 3) or dispatch_apply (in Swift 2):

DispatchQueue.concurrentPerform(iterations: iterations) { index in
    // Do something computationally expensive here
}

dispatch_apply(iterations, queue) { index in
    // Do something computationally expensive here
}

The loop closure will be invoked for each index from 0 to, but not including, iterations. These iterations will be run concurrently with respect to each other, and thus the order that they run is not guaranteed. The actual number of iterations that happen concurrently at any given time is generally dictated by the capabilities of the device in question (e.g. how many cores does the device have).

A couple of special considerations:

To run tasks on a dispatch queue, use the sync, async, and after methods.

To dispatch a task to a queue asynchronously:

let queue = DispatchQueue(label: "myQueueName")

queue.async {
    //do something
    
    DispatchQueue.main.async {
        //this will be called in main thread
        //any UI updates should be placed here
    }
}
// ... code here will execute immediately, before the task finished

To dispatch a task to a queue synchronously:

queue.sync {
    // Do some task
}
// ... code here will not execute until the task is finished

To dispatch a task to a queue after a certain number of seconds:

queue.asyncAfter(deadline: .now() + 3) {
    //this will be executed in a background-thread after 3 seconds
}
// ... code here will execute immediately, before the task finished

NOTE: Any updates of the user-interface should be called on the main thread! Make sure, that you put the code for UI updates inside DispatchQueue.main.async { ... }

Types of queue:

let mainQueue = dispatch_get_main_queue()
let highQueue = dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_HIGH, 0)
let backgroundQueue = dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_BACKGROUND, 0)

To dispatch a task to a queue asynchronously:

dispatch_async(queue) {
    // Your code run run asynchronously. Code is queued and executed 
    // at some point in the future.
}
// Code after the async block will execute immediately

To dispatch a task to a queue synchronously:

dispatch_sync(queue) {
    // Your sync code
}
// Code after the sync block will wait until the sync task finished

To dispatch a task to after a time interval (use NSEC_PER_SEC to convert seconds to nanoseconds):

dispatch_after(dispatch_time(DISPATCH_TIME_NOW, Int64(2.5 * Double(NSEC_PER_SEC))), dispatch_get_main_queue()) {
    // Code to be performed in 2.5 seconds here
}

To execute a task asynchronously and than update the UI:

dispatch_async(queue) {
    // Your time consuming code here
    dispatch_async(dispatch_get_main_queue()) {
        // Update the UI code 
    }
}

NOTE: Any updates of the user-interface should be called on the main thread! Make sure, that you put the code for UI updates inside dispatch_async(dispatch_get_main_queue()) { ... }

You can think of an OperationQueue as a line of tasks waiting to be executed. Unlike dispatch queues in GCD, operation queues are not FIFO (first-in-first-out). Instead, they execute tasks as soon as they are ready to be executed, as long as there are enough system resources to allow for it.

Get the main OperationQueue:

let mainQueue = OperationQueue.main

Create a custom OperationQueue:

let queue = OperationQueue()
queue.name = "My Queue"
queue.qualityOfService = .default

Quality of Service specifies the importance of the work, or how much the user is likely to be counting on immediate results from the task.

Add an Operation to an OperationQueue:

// An instance of some Operation subclass
let operation = BlockOperation {
    // perform task here
}

queue.addOperation(operation)

Add a block to an OperationQueue:

myQueue.addOperation {
    // some task
}

Add multiple Operations to an OperationQueue:

let operations = [Operation]()
// Fill array with Operations

myQueue.addOperation(operations)

Adjust how many Operations may be run concurrently within the queue:

myQueue.maxConcurrentOperationCount = 3 // 3 operations may execute at once

// Sets number of concurrent operations based on current system conditions
myQueue.maxConcurrentOperationCount = NSOperationQueueDefaultMaxConcurrentOperationCount

Suspending a queue will prevent it from starting the execution of any existing, unstarted operations or of any new operations added to the queue. The way to resume this queue is to set the isSuspended back to false:

myQueue.isSuspended = true

// Re-enable execution
myQueue.isSuspended = false

Suspending an OperationQueue does not stop or cancel operations that are already executing. One should only attempt suspending a queue that you created, not global queues or the main queue.

The Foundation framework provides the Operation type, which represents a high-level object that encapsulates a portion of work that may be executed on a queue. Not only does the queue coordinate the performance of those operations, but you can also establish dependencies between operations, create cancelable operations, constrain the degree of concurrency employed by the operation queue, etc.

Operations become ready to execute when all of its dependencies are finished executing. The isReady property then changes to true.

Create a simple non-concurrent Operation subclass:

class MyOperation: Operation {

    init(<parameters>) {
        // Do any setup work here
    }

    override func main() {
        // Perform the task
    }

}

class MyOperation: NSOperation {

    init(<parameters>) {
        // Do any setup work here
    }

    override func main() {
        // Perform the task
    }

}

Add an operation to an OperationQueue:

myQueue.addOperation(operation)

This will execute the operation concurrently on the queue.

Manage dependencies on an Operation.

Dependencies define other Operations that must execute on a queue before that Operation is considered ready to execute.

operation2.addDependency(operation1)

operation2.removeDependency(operation1)

Run an Operation without a queue:

  operation.start()

Dependencies will be ignored. If this is a concurrent operation, the task may still be executed concurrently if its start method offloads work to background queues.

Concurrent Operations.

If the task that an Operation is to perform is, itself, asynchronous, (e.g. a URLSession data task), you should implement the Operation as a concurrent operation. In this case, your isAsynchronous implementation should return true, you'd generally have start method that performs some setup, then calls its main method which actually executes the task.

When implementing an asynchronous Operation begins you must implement isExecuting, isFinished methods and KVO. So, when execution starts, isExecuting property changes to true. When an Operation finishes its task, isExecuting is set to false, and isFinished is set to true. If the operation it is cancelled both isCancelled and isFinished change to true. All of these properties are key-value observable.

Cancel an Operation.

Calling cancel simply changes the isCancelled property to true. To respond to cancellation from within your own Operation subclass, you should check the value of isCancelled at least periodically within main and respond appropriately.

operation.cancel()

Syntax