The HTTP_HOST is obtained from the HTTP request header and this is what the client actually used as "target host" of the request. The SERVER_NAME is defined in server config. Which one to use depends on what you need it for. You should now however realize that the one is a client-controlled value which may thus not be reliable for use in business logic and the other is a server-controlled value which is more reliable. You however need to ensure that the webserver in question has the SERVER_NAME correctly configured. Taking Apache HTTPD as an example, here's an extract from its documentation:

如果没有指定ServerName，那么服务器将尝试通过对IP地址执行反向查找来推断主机名。如果在ServerName中没有指定端口，那么服务器将使用传入请求的端口。为了获得最佳的可靠性和可预测性，您应该使用ServerName指令指定显式的主机名和端口。

Update: after checking the answer of Pekka on your question which contains a link to bobince's answer that PHP would always return HTTP_HOST's value for SERVER_NAME, which goes against my own PHP 4.x + Apache HTTPD 1.2.x experiences from a couple of years ago, I blew some dust from my current XAMPP environment on Windows XP (Apache HTTPD 2.2.1 with PHP 5.2.8), started it, created a PHP page which prints the both values, created a Java test application using URLConnection to modify the Host header and tests taught me that this is indeed (incorrectly) the case.

After first suspecting PHP and digging in some PHP bug reports regarding the subject, I learned that the root of the problem is in web server used, that it incorrectly returned HTTP Host header when SERVER_NAME was requested. So I dug into Apache HTTPD bug reports using various keywords regarding the subject and I finally found a related bug. This behaviour was introduced since around Apache HTTPD 1.3. You need to set UseCanonicalName directive to on in the <VirtualHost> entry of the ServerName in httpd.conf (also check the warning at the bottom of the document!).

<VirtualHost *>
    ServerName example.com
    UseCanonicalName on
</VirtualHost> 

这对我很管用。

总之，SERVER_NAME更可靠，但您依赖于服务器配置!

请注意，如果你想使用IPv6，你可能想使用HTTP_HOST而不是SERVER_NAME。如果输入http://[::1]/，则环境变量如下:

HTTP_HOST = [::1]
SERVER_NAME = ::1

这意味着，如果你做一个mod_rewrite，你可能会得到一个糟糕的结果。SSL重定向示例:

# SERVER_NAME will NOT work - Redirection to https://::1/
RewriteRule .* https://%{SERVER_NAME}/

# HTTP_HOST will work - Redirection to https://[::1]/
RewriteRule .* https://%{HTTP_HOST}/

这只适用于在没有主机名的情况下访问服务器。

我花了一段时间才理解人们所说的“SERVER_NAME更可靠”是什么意思。我使用共享服务器，不能访问虚拟主机指令。因此，我在.htaccess中使用mod_rewrite来将不同的HTTP_HOSTs映射到不同的目录。在这种情况下，HTTP_HOST是有意义的。

The situation is similar if one uses name-based virtual hosts: the ServerName directive within a virtual host simply says which hostname will be mapped to this virtual host. The bottom line is that, in both cases, the hostname provided by the client during the request (HTTP_HOST), must be matched with a name within the server, which is itself mapped to a directory. Whether the mapping is done with virtual host directives or with htaccess mod_rewrite rules is secondary here. In these cases, HTTP_HOST will be the same as SERVER_NAME. I am glad that Apache is configured that way.

但是，基于ip的虚拟主机的情况有所不同。在这种情况下，且仅在这种情况下，SERVER_NAME和HTTP_HOST可以是不同的，因为现在客户端通过IP而不是名称来选择服务器。确实，在某些特殊的构型中这很重要。

因此，从现在开始，我将使用SERVER_NAME，以防我的代码移植到这些特殊配置中。

正如balusC所说，SERVER_NAME是不可靠的，可以在apache配置中更改，服务器的服务器名配置以及您和服务器之间的防火墙。

以下函数总是返回没有端口的真实主机(用户键入的主机)，几乎是可靠的:

function getRealHost(){
   list($realHost,)=explode(':',$_SERVER['HTTP_HOST']);
   return $realHost;
}

假设你有一个简单的设置(CentOS 7, Apache 2.4。x和PHP 5.6.20)，只有一个网站(不假设虚拟主机)…

在PHP的意义上，$_SERVER['SERVER_NAME']是PHP在$_SERVER超全局中注册的一个元素，基于httpd.conf中的Apache配置(**ServerName** directive with UseCanonicalName on)(可能来自包含的虚拟主机配置文件，等等…)HTTP_HOST派生自HTTP主机头。将其视为用户输入。使用前进行筛选和验证。

下面是一个例子，我使用$_SERVER['SERVER_NAME']作为比较的基础。下面的方法来自我创建的名为ServerValidator (Validator的子类)的具体子类。ServerValidator在使用$_SERVER中的6或7个元素之前检查它们。

在确定HTTP请求是否是POST时，我使用了这个方法。

public function isPOST()
{
    return (($this->requestMethod === 'POST')    &&  // Ignore
            $this->hasTokenTimeLeft()            &&  // Ignore
            $this->hasSameGETandPOSTIdentities() &&  // Ingore
            ($this->httpHost === filter_input(INPUT_SERVER, 'SERVER_NAME')));
}

在调用此方法时，所有相关$_SERVER元素的过滤和验证(以及相关属性的设置)都已经完成。

台词……

($this->httpHost === filter_input(INPUT_SERVER, 'SERVER_NAME')

．.． 检查$_SERVER['HTTP_HOST']值(最终从请求的主机HTTP报头派生)是否与$_SERVER['SERVER_NAME']匹配。

现在，我使用超全局语言来解释我的示例，但这只是因为有些人不熟悉关于filter_input_array()的INPUT_GET、INPUT_POST和INPUT_SERVER。

The bottom line is, I do not handle POST requests on my server unless all four conditions are met. Hence, in terms of POST requests, failure to provide an HTTP host header (presence tested for earlier) spells doom for strict HTTP 1.0 browsers. Moreover, the requested host must match the value for ServerName in the httpd.conf, and, by extention, the value for $_SERVER('SERVER_NAME') in the $_SERVER superglobal. Again, I would be using INPUT_SERVER with the PHP filter functions, but you catch my drift.

请记住，Apache经常在标准重定向中使用ServerName(例如，将URL后面的斜杠去掉:例如，http://www.example.com变成http://www.example.com/)，即使您没有使用URL重写。

我使用$_SERVER['SERVER_NAME']作为标准，而不是$_SERVER['HTTP_HOST']。在这个问题上有很多争论。$_SERVER['HTTP_HOST']可以是空的，所以这不应该是创建代码约定的基础，比如我上面的公共方法。但是，仅仅因为它们都是集合并不能保证它们是相等的。测试是最好的方法(记住Apache版本和PHP版本)。

如果你想检查server.php或其他什么，你想用下面的方法调用它:

<?php
    phpinfo(INFO_VARIABLES);
?>

or

<?php
    header("Content-type: text/plain");

    print_r($_SERVER);
?>

然后使用您站点的所有有效url访问它，并检查差异。