| TCP Sockets Resource assignment issue [message #41059] |
Tue, 09 November 2010 22:00 |
k4nd4lf
Messages: 1
Registered: November 2010
|
Junior Member |
|
|
We're having some issues with the following bean counters: numtcpsock, numfile, tcprcvbug & tcpsndbuf
In each container we're running one nginx with 10 mongrels that are required to responde to a high number of requests per second (1000 as a minimum). Each request is quite optimized in order to reduce DB connections
The HN where OpenVZ is hosted is a <server-description>, and we have up to eight containers with the same or lower configuration as attached. Our goal was to setup each container with 2GB RAM and at least 2 CPU.
The configuration of each container was initially made using vzsplit, although we have increased some parameters as we keep getting errors.
The problem is the webservers running on these containers stop working because of the amount of file descriptors created when a lot of tcp connections are in TIME_WAIT state.
Does anyone have an idea which is the right value for these counters?
Thanx a lot in advance.
|
|
|
|
OpenVZ Forum
Members
Search
Help
Register
Login
Home
