Subject: Re: Mail list envelope sender address
To: None <current-users@NetBSD.ORG>
From: D. J. Bernstein <email@example.com>
Date: 11/26/1996 00:53:36
Perry Metzger writes:
> This is not the place for this discussion AT ALL.
When Erik Fair criticized qmail, and Rob Black plugged Exim, you
_didn't_ say ``Hey, this discussion doesn't belong on the NetBSD list.''
You said---showing your utter confusion about the concept of latency---
``Dan actually claims that opening three connections to one machine and
sending three copies of the same message when you want to deliver the
same message to three recipients is somehow a performance win.''
So it's okay for you to talk about the technical issue, but not me?
Again, I'll happily respond to your comments on comp.mail.misc if you
post them there.
> > He continues to claim that parallelism does not reduce latency.
> I never made such a claim.
Yes, you did.
Me, on DRUMS, 18 March 1996: ``Nobody has ever disputed that multiple
RCPTs are, in extreme circumstances, a win. But they are often a loss,
and occasionally a major loss. I object to DRUMS insisting that they
_always_ be used.''
You, on DRUMS, in response: ``They aren't a loss if the remote agent
doesn't try to check the receiving address. This should be obvious from
the fact that setting up every new TCP connection for a new SMTP takes
at least one RTT; therefore, what you are saving isn't RTTs.''
That's utter nonsense. The truth is that parallel RTTs overlap. Doing
two RTTs in parallel, instead of in serial, _does_ reduce latency. I'm
astounded that you're continuing to deny this.
Sick of sendmail? Don't get mad; get qmail. http://pobox.com/~djb/qmail.html