Load Limiting

Central mail machines often can be overloaded. The best solution is to dedicate a more powerful machine to handling mail, but the load almost always expands to consume whatever resources are allocated.

The sendmail program enables you to limit the load. The goal of load limiting is to prevent wasting time during loaded periods by attempting to deliver large messages, messages to many recipients, and messages to sites that have been down for a long time.

Use the Ox and OX options to limit the load caused by sendmail. The default is not to set any load limits if no options are used. Both these configuration options take an argument that is an integer load average. For example, if you specify Ox4 and OX8, then the x load limiting will be used when the load is above four, and the X load limiting will be used when the load is above eight. When the load is above the value specified in the X option, the SMTP server does not accept connections from the network (locally originated mail and other mail like UUCP are not affected). The x option has a more subtle effect, controlling whether messages are queued for later delivery or delivered immediately. The general idea is always to deliver "small" messages immediately and to defer "large" messages for delivery during off-peak periods.

The Oq option specifies the maximum size of a message that is delivered immediately. The "size" of the message includes not only the number of bytes in the message but also the penalty for a large number of recipients and for unsuccessful delivery attempts. The penalty per recipient is option value y, by default set to 1000. The penalty per delivery attempt is the option value z, by default set to 9000. The size limit also depends on the current load, so that more and more messages are queued as the load goes higher. If the load is one above the x threshold, then the limit is halved; if the load is two above the threshold, the limit is divided by three, and so on. Note that this limit also applies to messages being delivered when processing the queue.