并发是让两个任务在不同的线程上并行运行。然而,异步方法在同一个线程上并行运行。这是如何实现的?还有,并行性呢?

这三个概念有什么不同?


当前回答

Parallel : It's a broad term that means that two pieces of code execute that "at the same time". It doesn't matter if it's "real" parallelism or if it's faked through some clever design pattern. The point is that you can start the "tasks" at the same time and then control them separately (with mutex and all the appropriate tricks). But usually you prefer to use the word "parallel" only for "true" parallelism, as in : you make it happen through non-cooperative multitasking (whether be throuch CPU/GPU cores, or only at software level by letting the OS managing it at a very low level). People are reluctant to say "parallel" just for complicated sequential code that fakes parallelism, like you would find in a browser window's javascript for example. Hence the reason why people in this thread say "asynchronous has nothing to do with parallelism". Well it does, but just don't confuse them.

并发:没有并行性就不可能有并发性(无论是模拟的还是真实的,正如我上面解释的那样),但是这个术语特别关注的是两个系统将试图在某个时间点同时访问同一资源的事实。它强调了一个事实,那就是你必须要处理这个问题。

异步:每个人都说异步与并行无关,这是对的,但它为并行铺平了道路(让事情并行或不并行的负担在你身上——继续阅读)。

“异步”指的是并行性的一种表示形式,它形式化了并行性中通常涉及的三个基本内容:1)定义任务的初始化(比如它何时开始以及它获得哪些参数),2)任务完成后必须做什么,3)代码在此期间应该继续做什么。

但它仍然只是语法(通常表示为回调方法)。在后台,底层系统可能简单地认为这些所谓的“任务”只是堆积起来的代码片段,直到它完成当前正在执行的代码。然后它一个接一个地把它们拆开,然后依次执行。与否。它还可以为每个任务创建一个线程,并并行运行它们。谁在乎呢?这部分不包括在概念中;)

其他回答

这里有一些语义需要澄清:

并发或并行是资源争用的问题,而异步是关于控制流的问题。

不同的过程(或它们的组成操作)被称为异步的,当它们的处理顺序没有确定的实现时;换句话说,它们中的任何一个都有可能在任何给定的时间t被处理。根据定义,多个处理器(例如cpu或Persons)使多个处理器同时被处理成为可能;在单个处理器上,它们的处理是交错的(例如线程)。

当异步过程或操作共享资源时,它们被称为Concurrent;当没有资源共享(例如不同的处理器和存储)时,并行性是微不足道的保证;否则必须解决并发控制问题。

因此,异步过程或操作可以与其他过程或操作并行或并发地处理。

Parallel : It's a broad term that means that two pieces of code execute that "at the same time". It doesn't matter if it's "real" parallelism or if it's faked through some clever design pattern. The point is that you can start the "tasks" at the same time and then control them separately (with mutex and all the appropriate tricks). But usually you prefer to use the word "parallel" only for "true" parallelism, as in : you make it happen through non-cooperative multitasking (whether be throuch CPU/GPU cores, or only at software level by letting the OS managing it at a very low level). People are reluctant to say "parallel" just for complicated sequential code that fakes parallelism, like you would find in a browser window's javascript for example. Hence the reason why people in this thread say "asynchronous has nothing to do with parallelism". Well it does, but just don't confuse them.

并发:没有并行性就不可能有并发性(无论是模拟的还是真实的,正如我上面解释的那样),但是这个术语特别关注的是两个系统将试图在某个时间点同时访问同一资源的事实。它强调了一个事实,那就是你必须要处理这个问题。

异步:每个人都说异步与并行无关,这是对的,但它为并行铺平了道路(让事情并行或不并行的负担在你身上——继续阅读)。

“异步”指的是并行性的一种表示形式,它形式化了并行性中通常涉及的三个基本内容:1)定义任务的初始化(比如它何时开始以及它获得哪些参数),2)任务完成后必须做什么,3)代码在此期间应该继续做什么。

但它仍然只是语法(通常表示为回调方法)。在后台,底层系统可能简单地认为这些所谓的“任务”只是堆积起来的代码片段,直到它完成当前正在执行的代码。然后它一个接一个地把它们拆开,然后依次执行。与否。它还可以为每个任务创建一个线程,并并行运行它们。谁在乎呢?这部分不包括在概念中;)

每个人都很难将异步与并行或并发相关联,因为异步既不是并行的反义词,也不是并发的反义词。它是同步的反义词。它只是表示某些东西,在这个例子中是线程,是否会与其他东西同步,在这个例子中是另一个线程。

并发和并行实际上是相同的原理,两者都与同时执行的任务有关,尽管我想说并行任务应该是真正的多任务,“同时”执行,而并发可能意味着任务共享执行线程,同时仍然看起来是并行执行。

异步方法与前两个概念没有直接关系,异步被用来呈现并发或并行任务的印象,但实际上异步方法调用通常用于需要在当前应用程序之外执行工作的进程,我们不希望等待和阻塞应用程序等待响应。

例如,从数据库获取数据可能需要时间,但我们不想阻塞UI等待数据。异步调用接受回调引用,并在请求被放置到远程系统后立即将执行返回给您的代码。当远程系统执行所需的任何处理时,UI可以继续响应用户,一旦它将数据返回给回调方法,那么该方法就可以适当地更新UI(或移交更新)。

从用户的角度来看,它看起来像多任务处理,但它可能不是。


EDIT

可能值得补充的是,在许多实现中,异步方法调用会导致线程启动,但这不是必要的,这实际上取决于正在执行的操作以及如何将响应通知回系统。

当一个经理有几个员工,并且可以给他们每个人一个单独的任务时,并行性就会发生。工人们做他们的工作,并向经理提供结果。如果任务不能完全分离,例如任务之间的结果有一定的依赖性,或者需要在没有其他推理的情况下使用相同的资源,那么并行度就受到这些约束,不能完全实现。

Concurrency happens when a manager has several tasks but only less workers, hence some workers are given more than one task. Any worker given multiple tasks, divides each original given task to several steps and does the steps interleaved, each task result will be given back to manager as soon as every steps of it finished. Manager receive a task result while other tasks started and progressed several steps but have not finished yet. If any worker with multiple task decides not to start a single step of a given task before finishing every steps of an already started task, this is called sequentiality.

Asynchrony is any of the two above mixed or separated, seen from the manager's point of view. When the manager assigns the tasks to either few or enough workers he shall not be awaited stalled until any results are given back. He can do his personal jobs or whatever, while jobs are progressing. Usually workers do not decide how tasks should be divided into steps. An inversion of control means manager decides over steps and gives single steps to workers. So when he receives a step result from a worker, give him another step, maybe from another task. The whom under control is responsible for composing back step results into task results as well. So Asynchronicity comes with responsibility for control and probably coordination. If any worker is urged to work sequentially, from manager's point of view he is a synchronous worker.

Summary As it's simple to guess, full parallelism is an unrealisable idea unless otherwise in rare mostly trivial cases. Since reality comes with interdependent tasks and shared resources and lack of workers. So concurrency is the reality. From manager's point of view this concurrency is best if it does not hinder him from fine controlling the tasks, and if positive it is called asynchronous. Also computer software engineering best practices, augmented by S in SOLID principle, historically made servers single step runners called micro-services, this returned back control to the clients. So current situation is concurrency from server point of view and asynchronicity from client point of view.