Next-Hop SMTP Traffic Control

Ideally a mail server would deliver all mails to recipients' mailboxes immediately, but that is not always possible due to limited system resources. For active sites, mail volume can be very high. Clients sometimes attempt to deliver large quantities of mail in a very short time period, often mail which has low perceived value to the recipient. Those spikes in mail volume can exceed server capacity for short periods and delay processing of other mail with a higher perceived value. Such servers therefore attempt to throttle traffic from clients with a lower perceived value. Servers will also throttle or defer mail from senders for other reasons, including reputation or behaviour based spam filters that prefer to wait for more data before they make a delivery decision. There are two approaches currently used to throttle mail traffic. One is to limit the bandwidth of the transport the SMTP session is running over. By delaying the delivery of each message while keeping the SMTP session open it causes each connection to be held open for longer, significantly increasing resource usage at both ends of the connection. The other approach is to defer accepting a message with an RFC 5321 transient negative completion reply, typically "421", relying on the client to resend the message later. This can cause many repeated, failed delivery attempts when the client retries delivery of the same message rapidly, or attempted delivery of many (thousands) of messages, all of which are deferred. That pattern of high volumes of deferred connections can have an aggregate effect that is expensive and wasteful for both the client and server. Many clients will retry delivery only after a fairly long period (RFC 5321 suggests 30 minutes) which can cause mail delivery to be significantly delayed. Long delivery delays are an obvious problem for time-sensitive messages, bit they can also cause problems indirectly by causing messages to be delivered out of chronological order. If the server is only temporarily overloaded, and so needs to reduce traffic for just a short period (seconds to minutes), then a shorter retry delay reduces those problems.

This specification defines a new extended response that SMTP server can include in the text section of any SMTP response - separated by white space from the rest of the text section. It's intended use is to refine the second mechanism described above, to guide future transfer attempts by the client.

The delay value is the length of time, in seconds, that the server is asking the client to wait before sending further mail. A value of "0" is valid, with the meaning that the client SHOULD gracefully terminate the current connection, but need not wait before attempting future connections. Disconnecting in this way, rather than keeping the connection open for further messages, moves the client to the back of the connection queue and gives competing clients an opportunity to connect. If a client sees this additional response in any SMTP response from a server then it SHOULD gracefully close the SMTP connection by sending a QUIT command as soon as possible. It SHOULD NOT make any new connection to the same server for the duration of the delay. If possible it SHOULD gracefully close any other connections to the same server. If the implementation of the client makes it difficult to implement some of these behaviours, it can still usefully implement the other requirements. Other than this, it makes no changes in the semantics of the SMTP transaction.

A server can add a wait: additional response to any permanent or transient negative completion reply ("5xx or 4xx error") to ask the client to disconnect and wait before reconnecting. A server can also usefully add a wait: additional response to a 250 positive completion response at the end of the DATA section to ask the server to disconnect and wait before reconnecting. At that point in the transaction a "250 wait:" asks the client to disconnect rather than attempting to send further messages in the same transaction.

"421 4.3.2 wait:15 example.com Service not available" means the client should retry delivery of the mail, after at least 15 seconds and not attempt any other deliveries for at least 15 seconds. "554 wait:60 Transaction failed" means the client should not retry delivery of this message, and should not attempt any other deliveries for at least 60 seconds. "250 OK wait:90" in response to end of data means that the mail has been accepted for delivery and that the client should send a QUIT command, and not attempt delivery of any other messages for 90 seconds.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119.

This memo includes no request to IANA.

This document describes an additional response code. Use of this response code may disclose additional information about how a mail system is implemented beyond that currently available.