On Tue, 14 Jun 2005, Nigel Metheringham wrote:
> On Tue, 2005-06-14 at 12:20 +0100, Jonathan Knight wrote:
> > Does PIPELINING really improve the throughput?
>
> Probably only significantly if there are several recipients on a laggy
> connection - you save yourself a round trip time per recipient.
It's most significant for interactive message submission from
low-bandwidth high-RTT clients. With PIPELINING the time taken to send a
message is 1RTT for the TCP handshake, 1RTT for the server banner, 1RTT
for EHLO/response, 1RTT for the envelope, 1RTT for the data, and 1RTT for
QUIT. Without it the envelope takes at least 3RTT.
The LEMONADE people who have been focussing on email for limited clients
(e.g. phones) really screwed the pooch when they were defining enhanced
message submission. They should have turned it into an IMAP extension,
which would have made one of their goals MUCH simpler (to assemble a
message using attachments stored on the IMAP server without downloading
them and uploading them again). Instead they have defined a disgustingly
evil out-of-band channel between the SMTP server and the IMAP server
involving cookies in IMAP URLs, and they lost the opportunity to define a
message submission command that takes 1RTT total (assuming an existing
IMAP connection). </rant>
I don't think PIPELINING matters for MTA-MTA SMTP - the cost of spam is
far far worse than the RTT or packet header overhead!
Tony.
--
<fanf@???> <dot@???>
http://dotat.at/ ${sg{\N${sg{\
N\}{([^N]*)(.)(.)(.*)}{\$1\$3\$2\$1\$3\n\$2\$3\$4\$3\n\$3\$2\$4}}\
\N}{([^N]*)(.)(.)(.*)}{\$1\$3\$2\$1\$3\n\$2\$3\$4\$3\n\$3\$2\$4}}
