ietf-nntp Virtual hosts in NNTP servers

Joao Prado Maia jpm at papercut.org
Sat Mar 1 21:03:57 PST 2003


On Thu, 27 Feb 2003, Clive D. W. Feather wrote:

> If you want something that works now, then a protocol change isn't going 
> to work for you. In that case, you need to do something like playing 
> games with AUTHINFO, since clients already know how to do that.
> 
> Or do you need to provide both servers on port 119? Is the ability to 
> specify the port number unique to Turnpike or does everyone offer it? If 
> it is common, then why not just use different ports?
> 

I don't mind running the servers in separate ports, but I would rather 
have them all in port 119. And I'll tell you why - if you put the 
following in a web page:

news://servername.org/[newsgroup]

Then if the user is running Outlook Express, it will try to open the NNTP 
server automatically for you (creating the appropriate news account and 
etc), but it doesn't seem to recognize port numbers in the URL. That is, 
if you put:

news://servername.org:4444/[newsgroup]

Then Outlook Express will basically just create a news account for server 
'servername.org' with the title 'servername.org:4444'. That's basically 
how I got into trying to run both servers in port 119 and getting into the 
obvious conclusion that it wouldn't be possible.

Now, I understand that most news clients allow the user to change the port 
number for each news account, but again, we would get into the same 
problem as the AUTHINFO idea IMHO. I would have to explain to my users 
that they need to open dialog box XYZ and set the appropriate port number 
(since the port number in the URL doesn't work for the most popular 
newsreader around).


> Is this going to be part of the base protocol, or can it just be an 
> extension? Making it an extension means that it can be specified 
> separately and doesn't hold up the main document.
> 

I would prefer it to be part of the base protocol, but it seems like most 
people here are against that. It could work both ways, but having it in 
the document would help


> Will you expect all clients to always use HOST? If so, that's extra load 
> on the connection which 99% of users won't need. If not, how will they 
> know when to make the request? Will it be a flag that users have to set 
> or clear when configuring their newsreader? Will the server use a new 
> greeting code (e.g. 202, or 270) to indicate the need for HOST?
> 

How about making it optional (in a way) like HTTP does ? If the client 
doesn't send the HOST command, then we know that the client is trying to 
use the old version of the protocol.


> What should the server do if the client doesn't use HOST? If it uses it 
> with an unrecognised domain? What if it uses HOST after GROUP? What if 
> it uses it more than once during the session? Must it come before any 
> MODE READER, after it, or doesn't it matter?
> 
> What should the client do if the server responds 500 to HOST?
> 
> How much will any of the above constrain implementers unnecessarily?
> 

All of these questions are pretty good ones, but I think we should first 
agree in adding this new command or not, and then talk about its 
semantics.

Cheers,
Joao




More information about the ietf-nntp mailing list