I am surprised to learn that the default value for MaxClients on a Debian install is as low as 20. The value of MaxClients depends on a number of factors including the number of requests per second and the time required to serve each request. The longer it takes to serve a request, the longer a slot will be occupied and unavailable for serving other requests. The number of requests per second depends on the number of simultaneous users, characteristics of the web sites (number of graphical elements per page), user patterns... It also depends on the protocol used (HTTP/1.0 or 1.1), whether KeepAlive is used and the configuration of the browser (number of simultaneous connections to the same web site in HTTP/1.0). You basically just need to adapt the value of MaxClients to the needs of your site and the performance of your hardware and OS. With the prefork MPM one process is forked for each incoming connection and there is a need to limit the number of httpd processes on the system in order not to fill the process table and memory, and context-switching is expensive so performance becomes an issue. Using the worker MPM you can have one process handle just about as many connections as you wish (in my case 100) so processes and memory is a lesser preoccupation. Recent processors are also very good at doing multi-threading. If Apache does little work for each request but spends a lot of time waiting for a back-end to reply, then you can set MaxClients to a very high value. If the requests are processor-intensive you may need to reduce MaxClients and align several servers to handle the total volume of requests with reasonable performance. Then there may be considerations like the need to throttle the traffic to back-end application servers when the Apache server is used as a reverse proxy that will make you set a lower limit than you would otherwise. If you are suspecting a DoS attack you should start by looking for unusual activity in your server logs. I however find it very unlikely that it is the case. If you were under attack, raising the value to 100 would not have done the trick. -ascs -----Original Message----- From: dan+apache-list@xxxxxxxxxxxxxxx [mailto:dan+apache-list@xxxxxxxxxxxxxxx] Sent: Wednesday, February 08, 2006 7:11 AM To: users@xxxxxxxxxxxxxxxx Subject: [users@httpd] Why the sudden need to raise MaxClients? We are using Apache 2.0.54 on Debian GNU/Linux (standard Debian packages). The server had been running for several months with absolutely no problems. A couple of days ago, Apache started slowing down, then would eventually stop responding completely. The only way to fix it was to restart Apache. It got to the point that Apache would only stay functional for a few seconds at a time. Watching the Apache logs revealed no unusual activity before the lockup. I discovered the MaxClients setting, which in Debian defaults to 20 for the prefork MPM. I raised it to 100, and the problem went away. I believe the default for this value is normally 256. What would cause this problem all of the sudden? My first thought was a DOS attack. Am I on the right track? --df --------------------------------------------------------------------- The official User-To-User support forum of the Apache HTTP Server Project. See <URL:http://httpd.apache.org/userslist.html> for more info. To unsubscribe, e-mail: users-unsubscribe@xxxxxxxxxxxxxxxx " from the digest: users-digest-unsubscribe@xxxxxxxxxxxxxxxx For additional commands, e-mail: users-help@xxxxxxxxxxxxxxxx --------------------------------------------------------------------- The official User-To-User support forum of the Apache HTTP Server Project. See <URL:http://httpd.apache.org/userslist.html> for more info. To unsubscribe, e-mail: users-unsubscribe@xxxxxxxxxxxxxxxx " from the digest: users-digest-unsubscribe@xxxxxxxxxxxxxxxx For additional commands, e-mail: users-help@xxxxxxxxxxxxxxxx
|