Draft specification for future X-Trace header

Olaf Titz olaf at bigred.inka.de
Sat Jul 8 10:30:11 UTC 2000


> How common?  1% of all USENET users?  10,000 users?  15 users??
> I can't see this being as common as this discussion is making out.
> I fail to see why a customer would be running a news *server* on their home
> machine.  (But I *invite* people to educate me...)

Because their household consists of three people, all reading news
with a significant common subset of groups (but with different
newsreaders). Because they live in a country where ISP access is
charged by the minute and decidedly don't want to read news online. (I
remember a situation where I once ran up a monthly phone bill of $100
for the newsgroups and email read by _one person_ by _UUCP_.)

If you take the part of the population which knows how to run Linux
and INN, this configuration is _very_ popular in this high-price
country. Given 3 million private Internet users and 5% of them running
Linux I would estimate a potential of up to 50,000 home INN systems in
this country alone. (A significant portion of the remaining 100,000
are probably using leafnode which technically is a news server using a
POST feed too. The rest are using Netscape in offline mode.)

> I keep coming back to the idea that a news server doesn't belong in
> a customers house, and if it is there, it should abide by standard
> transfer protocol rather than bastardizing the standards by
> pretending (kind of) to be a reader. Remember that the overwhelming
> majority of client connections to the news server are *proper*
> newsreader connections.

The majority doesn't matter at all. If you buy the majority argument,
you can as well standardize that the only newsreader ever used should
be Outlook Express, and quoted articles should be appended uncut at
the bottom.

What does matter is that the network does work in a stable fashion, no
matter what quirky software people come up with next. As long as they
obey the simplest rules, like never changing any existing Message-ID
header, they can't cause disasters. But we have a long history of
gatewaying with Fidonet and similar networks, and those gateways have
frequently caused major havoc to Usenet proper by re-injecting old
articles munged by the gateways. (Split into 16kB parts, etc.)

Fidonet has its own history of being plagued by loops, and this was
caused primarily because it originally had no built-in stability and
relied on people to configure their systems correctly. Even when "the
overwhelming majority" did that right one malfunctioning node was
enough to flood the network with duplicates. They had to retrofit Path
and Message-ID headers to the message format to keep this under control.

> In your example case, arrange for the customer to have a static IP
> address and set them up as a proper newsfeed on your server.  (From

This is how in an ideal world that would work, without RIPE assignment
rules and understaffed mass-market ISPs with 2 million customers.
In reality, as a customer you have to take what the market gives.
"Abiding by the standard transfer protocol" is not possible for the
common citizen any more, period.

> the ISP's point of view, *if* I had a customer with the need, this
> would be my course of action.)

You obviously aren't T-Online (Germany's biggest ISP). It took several
months to convince T-Online to stop rewriting incoming Message-ID
headers, and it took several major duplicate floods which actually
happened. (At first they were even rewriting From headers. Do you want
to have the name of some random person appearing as the author of your
posts? I don't.) When you're as big as T-Online, you don't act upon
your users' needs and wishes but on something with is considered
standard by the "overwhelming majority"[1], and it is absolutely not
possible to configure _anything_ in your system for any user
individually.

Olaf

[1] In German they are called "WinDAUs". DAU="stupidest imaginable user".




More information about the inn-workers mailing list