Chuq Von Rospach <firstname.lastname@example.org> wrote:
> At 8:48 AM -0500 11/18/99, Tom Neff wrote:
> > The number of spams received PER USER
> > are no higher than any other ISP or service, big or small.
> that's probably not true, but beyond that, it's a purely theoretical
I maintain AOL accounts (along with accounts at Compuserve, Juno, Yahoo
etc - it's the best way to see what my users are seeing) and it does seem to
be true. I get spammed at AOL approx. 8 times per week, versus about 10-15
times per week at (for example) panix. The only accounts I have where spam
is nonexistent are at services like Bigfoot where they offer an aggressive
spam filter as a checkbox option.
So this is not a theoretical argument at all...
> that I don't think holds water. it's like saying that
> there's no real difference to being in a room with one
> non-housebroken dog and 20. In theory, you're right. In practice, you
> can bet there is.
That is something of a non sequitur. What I am saying is that the ability
to provide service, PER USER, should not degrade as an ISP grows, and if it
does, there is something wrong. People throw the "18 million spams/day"
number around as though this were some storefront operation in Muncie being
mailbombed by the big boys. AOL *is* the big boys! They make on the order
of $50 million PER MONTH. They have humongous data centers, and they buy
warp-speed mainframe and network hardware by the truckload. They can afford
to hire the most brilliant programmers and systems architects in the
country. If there is one organization in the world that ought to be able to
handle its mail properly, America Online is it.
The fact that they don't - that they drop legitimate mail on the floor while
continuing to allow spams to reach the user's mailbox - implies one or both
of the following: (1) they're not quite as smart about this as they could
afford to be; (2) scaling mail up to that level is a harder problem than
I suspect that the answer is a combination of the two. The question is how,
as list managers, we deal with it. I will write about that separately since
Chuq's digression makes for an unenlightening subthread.
> I find it fascinating to watch people tell other folks how easy it is
> to do their job right, myself.
We all find fascination where we can. I have been trying to figure out how
they make those flatbreads with the sesame seeds.
For the record, in saying that users have a right not to expect their
service level to degrade as their ISP grows in size, I was not telling
anyone how to do their job. Perhaps Chuq is responding to other messages,
or just reading lazily, I don't know.
> I wonder how many people on this list
> could begin to architect AOL's system, much less build one that
> didn't implode in the first three minutes.
This is a killer argument except for the likelihood that AOL's mail system
is not truly "architected" so much as grown organically, like the Mir space
station, over the years as AOL grew. Nobody out here has scratch-built a
mail system that's ready to serve 20 million users -- but nobody at AOL has
really done it either! They probably wish they could give it a shot.
Nevertheless, a few of us really have built some big systems, and there are
choices that you make. One is the software; AOL rolled their own, probably
after convincing themselves that this was the only way to handle the volume.
Another is the bastion server deployment; AOL relies on 20 Internet
machines, arranged using round-robin MX (four records) and then round-robin
DNS (five IPs per MX name). (Add to that an unknown number of machines
handling internal AOL-AOL traffic, probably using different software and
routing.) Assuming that most of the 18 million spams per day are from the
Internet, this means that each server receives around a million spams per
day. Still a big number, but these are very powerful machines. ...And so
forth, I don't want to think about it too deeply, I got my own garden to