Cost of a thread in C++ under Linux

栏目: IT技术 · 发布时间: 4年前

内容简介:Almost all our computers are made of several processing cores. Thus it can be efficient to “parallelize” expensive processing in a multicore manner. That is, instead of using a single core to do all of the work, you divide the work among multiple cores. A

Almost all our computers are made of several processing cores. Thus it can be efficient to “parallelize” expensive processing in a multicore manner. That is, instead of using a single core to do all of the work, you divide the work among multiple cores. A standard way to approach this problem is to create threads.

A C++ thread object executes some functions, possibly on a thread created by the operating system, and goes away. If you wanted to increment a counter using a C++ thread, you could do it in this manner:

auto mythread = std::thread([] { counter++; });
mythread.join();

It is obviously silly code that you should never use as is, it is a mere illustration. Creating a new thread is not free. Exactly how expensive it might be depends on several parameters. But can we get some rough idea?

For this purpose, I wrote a small benchmark where I just create a thread , increment a counter and let the thread die. It is the time elapsed while waiting for the thread to run its course. My program computes the mean and standard error of the time, as well as the minimum and maximum duration of the test. For simplicity, I am just going to report the means:

system time per thread
Ampere server (Linux, ARM) 200,000 ns
Skylake server (Linux, Intel) 9,000 ns
Rome server (Linux, AMD) 20,000 ns

I am deliberately not going into the details of the compiler, system library, operating system, RAM and all that fun stuff. You should not look at my table and make far reaching conclusions.

What is clear, however, is that creating a thread may cost thousands of CPU cycles. If you have a cheap function that requires only hundreds of cycles, it is almost surely wasteful to create a thread to execute it. The overhead alone is going to set you back.

There are clever ways to amortize the cost of a thread. For example, you may avoid the constant creation of new threads as in my benchmark. Yet to amortize, you still need to have enough work to make it worthwhile.


以上就是本文的全部内容,希望本文的内容对大家的学习或者工作能带来一定的帮助,也希望大家多多支持 码农网

查看所有标签

猜你喜欢:

本站部分资源来源于网络,本站转载出于传递更多信息之目的,版权归原作者或者来源机构所有,如转载稿涉及版权问题,请联系我们

密码学原理与实践

密码学原理与实践

[加]Douglas R.Stinson / 冯登国 / 电子工业出版社 / 2009年 / 55.00元

冯登国(FENG Dengguo,1965.5~), 现为中国科学院软件所研究员、博士生导师,信息安全国家重点实验室主任,国家计算机网络入侵防范中心主任,国家信息化专家咨询委员会委员。目前主要从事信息与网络安全方面的研究与开发工作。一起来看看 《密码学原理与实践》 这本书的介绍吧!

JSON 在线解析
JSON 在线解析

在线 JSON 格式化工具

RGB CMYK 转换工具
RGB CMYK 转换工具

RGB CMYK 互转工具