In
<Pine(_dot_)LNX(_dot_)4(_dot_)62(_dot_)0507271114330(_dot_)25684(_at_)sokol(_dot_)elan(_dot_)net>
"william(at)elan.net" <william(_at_)elan(_dot_)net> writes:
On Tue, 26 Jul 2005, wayne wrote:
What are your reasons for considering HTTP to be heavy-weight protocol?
I dunno about Mike, but I consider HTTP to be heavy-weight because an
HTTP GET transaction will likely be as costly in terms of both
bandwidth and latency as an SMTP transaction.
HTTP GET is one transaction, SMTP is multiple ones.
I'm not sure what you mean by "transaction".
It is easy to miss the round-trip packet exchanges of the TCP setup
and teardown. Also, packet overhead makes the difference in sizes
between the key (via HTTP) and an email body (via SMTP) much less
significant. Yes, the minimal SMTP session is going to be slightly
more than the minimal HTTP session, but either is going to be much
more similar (and more expensive) than a DNS lookup.
I also think that there will be resistance by many mail admins and MTA
authors to adding in complete support for HTTP on their mail boxes.
How many unix systems do you know that don't have wget or equivalent?
There is a difference between having HTTP support on the same box and
HTTP support in the MTA. I, personally, don't have a huge problem
with HTTP support in the MTA, but what I'm pointing out is that I
think that there will be a resistance from many mail admins. Even if
their objections are unfounded and irrational, I think it is important
to consider the deployment problems of convincing mail admins that it
is ok. That, of course, is assuming that my guess about many mail
admins resisting such things is correct.
Do you also consider SMTP or LDAP to be heavy-weight?
That depends on the context. Many people also consider SMTP
call-backs for SMTP transactions to be too heavy-weight.
I don't think SMTP call-backs is quite the same thing. To be fair,
I also don't think its quite that heavy-weight either (and many
do use it and it works for them, so clearly it can be used).
Again, I'm pointing out my understanding of what other mail admins
think, not neccessarily my own opinions. In fact, I used to use SMTP
call-backs until I got too fed up with Yahoo's broken email.
SMTP call-backs don't need to do the DATA step, so that removes one or
two round trips, leaving just an HELO round trip, plus a MAIL
FROM/RCPT TO/QUIT packet. That is at most one round trip more
expensive than an HTTP request.
Maybe we should dig out tcpdump and actually measure things. (I don't
have time right now, so if someone who supports HTTP lookups could do
it, I would appreciate it.)
-wayne