Date: 6 Mar 1996 03:18:33 GMT
Message-ID: <19960306031833.11466.qmail@koobera.math.uic.edu>
From: djb@koobera.math.uic.edu (D. J. Bernstein)
To: drums@cs.utk.edu
Subject: Re: re PIPELINING

> It is important that "efficiency" not be confused with "real time
> performance."  These are two very different things.

Bullshit. Low latency---i.e., speed, from the user's point of view---
is most certainly a form of efficiency. It is dishonest to refer to
something as an ``efficiency feature'' when it SLOWS DOWN DELIVERY.

> This is not a new observation.  

Of course it's not new. It is, however, obviously not understood by the
people who wrote the relevant paragraph in RFC 1123 and the people who
uncritically copied it into the DRUMS spec.

Here is a proposed rewrite of the paragraph:

   When a client sends a single message to two or more recipients over
   a single SMTP connection, it SHOULD NOT use the sequence MAIL, RCPT,
   DATA, MAIL, RCPT, DATA. The sequence MAIL, RCPT, RCPT, DATA uses less
   time and bandwidth.

This leaves parallel mailers free to respect the user's desired balance
between latency and network use.

> Nor are Bernstein's florid hyperbole,
> overstated (and incorrect) conclusions, and important omissions.

If I've said anything incorrect, quote it.

> This last is rather important; there are some non-obvious scaling and
> denial of service issues associated with not aggregating multiple

Bullshit. The issues are (1) obvious and (2) already solved.

---Dan