ietf-smtp
[Top] [All Lists]

Re: Max RCPT TO: - option to negotiate?

1998-04-05 17:23:55
And if you don't see the advantage, well, I would suggest that you do some
additional timing of SMTP behavior on high latency links. Pipelining often
makes a significant difference.

It will clearly make a big improvement on latency.  How much does your
experience suggest is helps on throughput?

FIguring how throughput is affected is much more difficult. It depends very
much on an MTA's queue processing strategy and the fan-out of message
destinations. For example, an MTA that enforces a small number of connections
per destination handling traffic to a small number of destinations is likely to
win big with pipelining. On the other hand, an MTA that doesn't pay attention
to destinations and fires up lots of connections to a large number of
destinations won't benefit much. But of course one can also argue that both of
these MTA strategies have major problems with certain kinds of loads.

In practice our customers tune things to process messages based on the mix of
traffic they handle, not on the basis of making pipelining work better. In most
cases we see pipelining give some benefit; it is rare for it to have no effect
and equally rare for it to win big.

                                Ned

<Prev in Thread] Current Thread [Next in Thread>