提升.NET 4.5中TPL性能的实战指南

需积分: 9 4 浏览量更新于2024-07-21 收藏 664KB PDF 举报

在.NET 4.5中，Task Parallel Library (TPL) 的性能改进是开发人员关注的重要话题。该篇文章由Joseph E. Hoag撰写，主要针对C#和VB.NET开发者，旨在提供一种相对轻松的方式来利用多线程和异步编程的能力。TLP的核心目标是简化并行编程，它分为两个关键部分： 1. 低级Task库：这一部分提供了极高的灵活性，允许开发者根据需要创建和配置各种并行或异步场景。用户可以自定义任务的执行逻辑，包括设置任务的优先级、依赖关系和错误处理等。这使得开发者能够更好地控制并发操作的细节。 2. 高级功能：基于Task的库如`Parallel.For`、`Parallel.ForEach`和`Parallel.Invoke`，这些方法内置了常见的并行编程模式，如循环迭代和并发执行多个操作，无需开发者从头开始编写复杂的同步代码。这些工具简化了诸如数据并行处理、任务分发和结果聚合等工作。在.NET 4版本中，TLP主要聚焦于 fork-join（任务拆分与合并）风格的编程，通过启动多个任务并等待它们全部完成来实现。例如，代码片段展示了如何使用`Task.Factory.StartNew`方法启动两个独立的任务，并通过`Task.WaitAll`确保它们都执行完毕。然而，在.NET 4.5中，TLP引入了更少结构化的并行形式，包括对延续支持的增强。这种编程方式避免了显式等待任务，而是通过指定任务之间的依赖关系，当一个任务完成后，自动执行后续的任务。这种方式有助于减少阻塞和提高代码的响应性，从而提升了整个应用程序的性能。 TLP在.NET 4.5中的性能改进着重于提供更灵活、易于理解和使用的并行编程工具，帮助开发者更有效地利用多核处理器，降低编程复杂性，提升应用程序的吞吐量和效率。通过学习和实践这些新特性，开发者可以更好地利用现代硬件，优化他们的.NET应用程序。

myEvent.Set(true);

}, event);

return event.Wait();

}

The upshot here is that, in most cases, there is no need to store a ManualResetEventSlim on the Task at

all; it can be “pay for play” for those wishing to use Wait() and the fork-join model. There is one

scenario, though, in which the m_completionEvent field is still needed – synchronously waiting on an

IAsyncResult, as in this example:

Task t1 = Task.Factory.StartNew( () => {});

((IAsyncResult)t1).AsyncWaitHandle.WaitOne();

In the case above, we need to instantiate/inflate the internal m_completionEvent, which in .NET 4.5

means inflating the Task’s ContingentProperties. However, the

IAsyncResult.AsyncWaitHandle.WaitOne() scenario is relatively rare, even amongst Asynchronous

Programming Model (APM) users, where callbacks are highly preferred to synchronous blocking.

Further, the cost of allocating a kernel synchronization primitive (as is done for ManualResetEvent) is a

much more significant cost than inflating contingent properties.

Finally, we moved Task.ContingentProperties.m_continuationList to Task.m_continuationObject in

order to optimize the performance of attaching continuations. This will be discussed in more detail in

the next section.

Continuation streamlining

In .NET 4, continuations were optimized as a secondary use case for Tasks, so the Task’s continuation list

was stored in Task.ContingentProperties. In .NET 4.5, continuations were optimized as a primary use

case for Tasks, due in no small part to the emergence of the async/await functionality which depends

heavily upon continuations.

Attaching the initial continuation for a Task in .NET 4 meant (1) inflating the Task’s

ContingentProperties, (2) allocating a continuation list, and (3) inserting the continuation into that list.

This was a pretty high price to pay (in terms of both time and memory) for adding a single continuation

to a Task, especially considering that analysis of common scenarios indicates that having at most one

continuation per Task is by far the most common case.

In .NET 4.5, we removed m_continuationList from Task.ContingentProperties, and replaced it with

Task.m_continuationObject. The m_continuationObject field can be null (meaning “no continuations”),

a single continuation object, a list of continuation objects, or a sentinel value meaning “this Task has

completed.” Attaching the initial continuation for a Task now requires a simple

Interlocked.CompareExchange operation, which is much faster than what we did in .NET 4. Attaching a

剩余19页未读，继续阅读

whl7711

粉丝: 0
资源: 11

提升.NET 4.5中TPL性能的实战指南

.net4.5离线安装包_x86_x64

Pro WPF 4.5 in C# WPF 编程宝典 2012和.NET4.5 第四版

WPF在.net4.0 无法使用 async await

.net 多线程高并发

.net core 多线程

.net 怎么用TPL开启多线程处理大量数据

.net3.5 实例源码

Write a function named sum_elements to compute the sum of all the elements of each tuple stored inside a list of tuples and return a list of these sums. The number of elements in the tuples might vary. For example, if lst_tpl = [(1,2,3),(5,6),(7,)], sum_elements(lst_tpl) should return [6, 11, 7]

.net 实现并行开发 代码

.net 加快查询速度

最新资源

.net 实现并行开发代码