DNS Prefetching

Problem

DNS resolution time can lead to a significant amount of user perceived latency. The time that DNS resolution takes is highly variable.  Latency delays range from around 1ms (locally cached results) to commonly reported times of several seconds.

Solution

DNS prefetching is an attempt to resolve domain names before a user tries to follow a link. This is done using the computer's normal DNS resolution mechanism; no connection to Google is used. Once a domain name has been resolved, if the user does navigate to that domain, there will be no effective delay due to DNS resolution time.  The most obvious example where DNS prefetching can help is when a user is looking at a page with many links to various domains, such as a search results page.  When we encounter hyperlinks in pages, we extract the domain name from each one and resolving each domain to an IP address.  All this work is done in parallel with the user's reading of the page, using minimal CPU and network resources.  When a user clicks on any of these pre-resolved names, they will on average save about 200 milliseconds in their navigation (assuming the user hadn't already visited the domain recently). More importantly than the average savings, users won't tend to experience the "worst case" delays for DNS resolution, which are regularly over 1 second.

Architecture

Chromium's implementation of DNS prefetching does not use the browser's network stack at all.  Instead, it relies on external threads to resolve the names, thereby warming the DNS cache of the operating system (completely ignoring any cache in the application network stack).  The advantage of this approach was that it was completely compatible with all network stacks (it is external), and prevented accidental regressions on the main network stack.

 
 
Since some DNS resolutions can take a long time, it is paramount that such delays in one resolution should not cause delays in other resolutions.  Toward this end (on Windows, where there is no native support for asynchronous DNS resolution), Chromium currently employs 8 completely asynchronous worker threads to do nothing but perform DNS prefetch resolution. Each worker thread simply waits on a queue, gets the next requested domain name, and then blocks on a synchronous Windows resolution function.   Eventually the operating system responds with a DNS resolution, the thread then discards it (leaving the OS cache warmed!), and waits for the next prefetch request.  With 8 threads, it is rare than more than one or two threads will block extensively, and most resolution proceed rather quickly (or as quickly as DNS can service them!).  On Debug builds, the "about:histograms/DNS.PrefetchQueue" has current stats on the queueing delay.

Manual Prefetch

Chromium uses the "href" attribute of hyperlinks to find host names to prefetch. However, some of those hyperlinks may be redirects, for example if the site is trying to count how many times the link is clicked.  In those situations, the "true" targeted domain is not necessarily discernible by examining the content of a web page, and so Chromium not able to prefetch the final targeted domain.

To improve the speed of redirects, content authors can add the following tag to their page:
 
<link rel="dns-prefetch" href="//host_name_to_prefetch.com">
 
The above "link rel" tag has no impact on the visual rendering of the page, but causes Chromium to prefetch the DNS resolution of "host_name_to_prefetch.com" as though there was an actual href targeted at a path in that domain. The double slashes indicate that the URL starts with a host name (as specified in RFC 1808). It is equivalent (but unnecessary) to use a full URL such as "http://host_name_to_prefetch.com/".

DNS Prefetch Control

By default, Chromium does not prefetch host names in hyperlinks that appear in HTTPS pages. This restriction helps prevent an eavesdropper from inferring the host names of hyperlinks that appear in HTTPS pages based on DNS prefetch traffic. The one exception is that Chromium may periodically re-resolve the domain of the HTTPS page itself.

An inquisitive content author (for example, a commenter on a blog) may abuse DNS prefetching to attempt to monitor viewing of content containing links. For example, links with novel subdomains, when resolved during a prefetch, may notify a domain's resolver that a link was viewed, even if it was not clicked.  In some such cases, the authority serving the content (such as a blog owner, or webmail server) may wish to preclude such abusive monitoring.

To allow webmasters to control whether DNS prefetch is enabled or disabled, Chromium includes a DNS Prefetch Control mechanism. It can be used to turn DNS prefetch on for HTTPS pages, or turn it off for HTTP pages.

Chromium watches for an HTTP header of the form "X-DNS-Prefetch-Control" (case insensitive) with a value of either "on" or "off."  This setting changes the default behavior for the rendered content. A "meta http-equiv" tag of the same name can be used to make policy changes within a page. If a page explicitly opts out of DNS prefetch, further attempts to opt in are ignored.
 
For example, the following page from https://content_author.com/ would cause Chromium to prefetch "b.com" but not "a.com", "c.com", or "d.com".

<a href="http://a.com"> A) Default HTTPS: No prefetching </a>
<meta http-equiv="x-dns-prefetch-control" content="on">
<a href="http://b.com"> B) Manual opt-in: Prefetch domain resolution. </a>
<meta http-equiv="x-dns-prefetch-control" content="off">
<a href="http://c.com"> C) Manual opt-out: Don't prefetch domain resolution </a>
<meta http-equiv="x-dns-prefetch-control" content="on">
<a href="http://d.com"> D) Already opted out: Don't prefetch domain resolution. </a>

Child frames also inherit the DNS prefetch control opt-out setting from their parent. The DNS prefetch control setting applies only to hyperlinks and not to the manual prefetch mechanism.

Browser Startup

Chromium automatically remembers the first 10 domains that were resolved the last time the Chromium was started, and automatically starts to resolve these names very early in the startup process.  As a result, the domains for a user's home page(s), along with any embedded domains (or anything the user "always" visits just after startup), are generally resolved before much of Chromium has ever loaded.  When Chromium finally starts to try to load and render those pages, there is typically no DNS induced latency, and the application effectively "starts up" (becoming usable) faster.  Average startup savings are 200ms or more, with common acceleration over 1 second.

Omnibox

Prefetching is also used in Chromium's omnibox, where URL and/or search queries are entered. The omnibox automatically proposes an action, either of the form of a search query, or a URL navigation, as the user types in text.  This proposed action is considered the autocompletion of what the user is typing.  Each time the omnibox makes a proposal (suggests an autocompletion), the domain for the underlying URL is automatically pre-resolved.  This means that when a user is entering a search query, while they type the query (typically when they enter a space between words), Chromium will automatically prefetch the resolution of the domain in their search provider's URL.  Assuming they haven't done a search in a while, this can save a user an average of over 100ms in getting a search result, if not more. Similarly, if the URL is a "common" URL that they've typed in the past (e.g., their favorite web site; bank;  email provider; their company; etc.), the savings in resolution time can be quite significant.

Effectiveness

Typical Usage

If a user resolved a domain name to an IP address recently, their operating system will remember (cache) the answer, and then resolution time can be as low as 0-1ms (a thousandth of a second).  If the resolution is not locally cached and needs to "go out over the network," then a bare minimum for resolution time is about 15 ms, assuming a nearby firewall (home router?) has a cached answer to the question.  Most common names like google.com and yahoo.com are resolved so often that most local ISP's name resolvers can answer in closer to 80-120ms.  If the domain name in question is an uncommon name, then a query may have to go through numerous resolvers up and down the hierarchy, and the delay can average closer to 200-300ms.  More interestingly, for any of these queries that access the internet, dropped packets, and overworked (under provisioned) name resolvers, regularly increases the total resolution time to between 1 and 10 seconds.

Network Usage

DNS resolutions use a hierarchical system, where each level in the hierarchy has a cache, to remember previous resolutions.  As a result, extra resolutions don't generally speaking cause end-to-end internet usage.  The resolutions only go as far as needed toward the "authoritative" resolver, stopping when they reach a resolver that already has the resolution in a cache.  In addition, DNS resolution requests are very light weight.  Each request typically involves sending a single UDP packet that is under 100 bytes out, and getting back a response that is around 100 bytes.  This minimal impact on network usage is compensated by a significant improvement in user experience.

Cache Eviction

The local machine's DNS cache is pretty limited.  Current estimates for the number of resolutions remembered on a Windows XP box are in the range of 50-200 domain names.  As a result, if "too many" resolutions are made, then some "necessary" resolutions might be "evicted" from the cache to make room for the new prefetches.  Chromium tries to model the underlying cache, and guess when there is a chance that a "soon to be needed" domain resolution has been evicted.  When Chromium decides an eviction may have taken place, it can automatically resolve the domain name again, ensuring it is either reloaded into the cache, or marked as "recently used" so that it won't be evicted for a "while."  Cache evictions caused by Chromium can have a negative impact on other applications that don't (yet) use prefetching techniques to keep the underlying cache "warm."

 
Many large internet sites (google.com, yahoo.com, etc.) commonly mark the domain names resolutions with an expiration time in the neighborhood of 5 minutes. They probably set the expiration time to be short so that they can better respond to changes in supply and demand for their services.  This in turn tends cause cache evictions.  As a result, most applications tend to face the problem of cache eviction, and already employ various methods to reduce the impact.
 
The "about:histograms/DNS" and "about:dns" pages contain more information about DNS prefetch activity.

DNS Prefetching的更多相关文章

  1. seo网页加速技术,预加载 DNS Prefetching 详解

    seo网页加速技术,预加载 DNS Prefetching 详解 DNS Prefetching 是什么 : DNS 是什么-- Domain Name System,域名系统,作为域名和IP地址相互 ...

  2. Controlling DNS prefetching

    Controlling DNS prefetching IN THIS ARTICLE Background Configuring prefetching in the browser Contro ...

  3. 前端优化:DNS预解析提升页面速度

    在网页体验中我们常会遇到这种情况,即在调用百度联盟.谷歌联盟以及当前网页所在域名外的域名文件时会遇到请求延时非常严重的情况.那么有没有方法去解决这种请求严重延时的现象呢? 一般来说这种延时的原因不会是 ...

  4. 【转】浏览器DNS 预取读技术的危害

    今天中午在http://news.ycombinator.com/news看到一篇文章标题: Saved 10 billion DNS queries per month by disabling D ...

  5. DNS Prefetch初认识

    今天在看一个后台框架时,发现这样的代码: <link rel="dns-prefetch" href="//0.s3.envato.com">< ...

  6. DNS 预解析

    DNS 解析也是需要时间的,可以通过预解析的方式来预先获得域名所对应的 IP. <link rel="dns-prefetch" href="//yuchengka ...

  7. DNS预解析dns-prefetch提升页面载入速度优化前端性能

    当浏览器请求一个URL的时候,通过firebug我们可以发现大概有以下几个过程:阻挡.域名解析.建立连接.发送请求.等待响应.接收数据.后面四个跟用户的网络情况和你的服务器处理速度有关,本文重点说说前 ...

  8. 前端优化系列之一:dns预获取 dns-prefetch 提升页面载入速度

    问题:怎么做到dns域解析?   用于优化网站页面的图片   问题:怎么提升网站性能? dns域解析,是提升网站的一个办法.   DNS Prefetch,即DNS预获取,是前端优化的一部分. 一般来 ...

  9. DNS预解析 dns-prefetch

    1.DNS 是什么? Domain Name System,域名系统,作为域名和IP地址相互映射的一个分布式数据库. DNS大家都懂,那么浏览器访问域名的时候,是需要去解析一次DNS,也就是把域名 g ...

随机推荐

  1. [Android] Android开发优化之——对界面UI的优化(2)

    在Android应用开发过程中,屏幕上控件的布局代码和程序的逻辑代码通常是分开的.界面的布局代码是放在一个独立的xml文件中的,这个文件里面是树型组织的,控制着页面的布局.通常,在这个页面中会用到很多 ...

  2. bzoj4519: [Cqoi2016]不同的最小割(分治最小割)

    4519: [Cqoi2016]不同的最小割 题目:传送门 题解: 同BZOJ 2229 基本一样的题目啊,就最后用set记录一下就ok 代码: #include<cstdio> #inc ...

  3. cximage功能简介

    CxImage是一个可以用于MFC 的C++图像处理类库类,它可以打开,保存,显示,转换各种常见格式的图像文件,比如BMP, JPEG, GIF, PNG, TIFF, MNG, ICO, PCX, ...

  4. 小型ceph集群的搭建

    了解ceph DFS(distributed file system)分布式存储系统,指文件系统管理的物理存储资源,不一定直接连接在本地节点上,而是通过计算机网络与节点相连,众多类别中,ceph是当下 ...

  5. Django——Model

    一. ORM 在 MVC 或者说 MTV 设计模式中,模型(M)代表对数据库的操作.那么如何操作数据库呢? 我们可以在 Python 代码中嵌入 SQL 语句. 但是问题又来了,Python 怎么连接 ...

  6. vector迭代器

    https://www.cnblogs.com/quant-lee/p/6618829.html

  7. 洛谷P1136 迎接仪式 动态规划

    显然,这是一道动归题. 我们发现,每次交换时只可能交换不同的字母(交换同类字母显然是没有意义的).那么每次交换等同于将 111 个 "j""j""j& ...

  8. Java线程之基础

    Java内存模型(jmm) 线程通信 消息传递 重排序 顺序一致性 Happens-Before As-If-Serial 一.线程的生命周期及五种基本状态 线程生命周期:新建.就绪.运行.阻塞.死亡 ...

  9. [luogu]P4364 [九省联考2018]IIIDX

    题目背景 Osu 听过没?那是Konano 最喜欢的一款音乐游戏,而他的梦想就是有一天自己也能做个独特酷炫的音乐游戏.现在,他在世界知名游戏公司KONMAI 内工作,离他的梦想也越来越近了. 这款音乐 ...

  10. C语言中头文件尖括号和引号的区别

    用include 引用头文件时,双引号和尖括号的区别: 1.双引号:引用非标准库的头文件,编译器首先在程序源文件所在目录查找,如果未找到,则去系统默认目录查找,通常用于引用用户自定义的头文件. 2.尖 ...