linux最大文件数限制

http://www.xenoclast.org/doc/benchmark/HTTP-benchmarking-HOWTO/node7.html

Performance tuning

General Tuning Tips
The essentials of tuning Apache for optimum performance are discussed in depth at
http://httpd.apache.org/docs/misc/perf-tuning.html. Some of the advice given here is definitely Apache specific, but parts (or at least the principles behind them) are more generally applicable.

Some simple rules follow:


Only enable those modules you absolutely require. Each additional module implies additional processing overhead, even though you may not be making active use of the module on your site.

Where possible, use static content instead of dynamic content. If you are generating weather reports that are updated once an hour, then it is better to write a program that generates a static file once an hour than to have users run a CGI to generate the report on the fly.

When choosing an API for your dynamic applications, choose the fastest and most appropriate one available. CGI may be easy to program for, but it forks a process for each request - usually an expensive and unnecessary procedure. FastCGI is a better alternative, as is Apache's mod_perl- both provide persistency and can increase performance significantly.

If your server performance is a critical issue for you (or you just want to get the maximum number of bangs for your bucks), then choose a high performance web server instead of Apache. Apache's process model sucks, frankly; it uses an individual process for each connection, leading to massive context switching when you have a large number of simultaneous connections. Single process per CPU servers such as Zeus and Boa are significantly faster; Zeus has held the SPECweb96 and 99 records for the majority of the last five years. Unfortunately, however, it is commercial, but is worth the money if you want to eke the last drop of performance out of your server.
Zeus: http://www.zeus.com/
Boa: http://www.boa.org/


Tuning file descriptor limits on Linux
Linux limits the number of file descriptors that any one process may open; the default limits are 1024 per process. These limits can prevent optimum performance of both benchmarking clients (such as httperf and apachebench) and of the web servers themselves (Apache is not affected, since it uses a process per connection, but single process web servers such as Zeus use a file descriptor per connection, and so can easily fall foul of the default limit).

The open file limit is one of the limits that can be tuned with the ulimit command. The command ulimit -aS displays the current limit, and ulimit -aH displays the hard limit (above which the limit cannot be increased without tuning kernel parameters in /proc).

The following is an example of the output of ulimit -aH. You can see that the current shell (and its children) is restricted to 1024 open file descriptors.


core file size (blocks) unlimiteddata seg size (kbytes) unlimitedfile size (blocks) unlimitedmax locked memory (kbytes) unlimitedmax memory size (kbytes) unlimitedopen files 1024pipe size (512 bytes) 8stack size (kbytes) unlimitedcpu time (seconds) unlimitedmax user processes 4094virtual memory (kbytes) unlimited

Increasing the file descriptor limit
The file descriptor limit can be increased using the following procedure:


Edit /etc/security/limits.conf and add the lines:
* soft nofile 1024* hard nofile 65535

Edit /etc/pam.d/login, adding the line:
session required /lib/security/pam_limits.so

The system file descriptor limit is set in /proc/sys/fs/file-max. The following command will increase the limit to 65535:
echo 65535 > /proc/sys/fs/file-max

You should then be able to increase the file descriptor limits using:
ulimit -n unlimited
The above command will set the limits to the hard limit specified in /etc/security/limits.conf.
Note that you may need to log out and back in again before the changes take effect.

文章来自: 本站原创
标签: linux 最大文件数 限制 ulimit