0.99.0 and later
The tcpBackLog directive controls the tcp "backlog queue" when listening for connections in standalone mode (see ServerType). It has no affect upon servers in inetd mode. When a tcp connection is established by the tcp/ip stack inside the kernel, there is a short period of time between the actual establishment of the connection and the acceptance of the connection by a user-space program. The duration of this latency period is widely variable, and can depend upon several factors (hardware, system load, etc). During this period tcp connections cannot be accepted, as the port that was previously "listening" has become filled with the new connection. Under heavy connection load this can result in occasional (or even frequent!) "connection refused" messages returned to the incoming client, even when there is a service available to handle requests. To eliminate this problem, most modern tcp/ip stacks implement a "backlog queue" which is simply a pre-allocation of resources necessary to handle backlog-size connections during the latency period. The larger the backlog queue, the more connections can be established in a very short time period. The trade-off, of course, is kernel memory and/or other kernel resources. Generally it is not necessary to use a tcpBackLog directive, unless you intend to service a large number of virtual hosts (see <VirtualHost>), or have a consistently heavy system load. If you begin to notice or hear of "connection refused" messages from remote clients, try setting a slightly higher value to this directive.