Re: Extended RETR - version 2

> 
> : Sorry, but I'm a bit doubtful about this extension. How serious
> : is the problem with a "noisy phoneline" really ? I'm just a naive
> : user (i.e not running an ISP) but I have never experienced this
> : problem, neither when living in Sweden nor in Australia.
> 
> A quick grep over the logfiles of a large ISP (>1 Mio POP accesses a day)
> tells me that less then 0.1% of the connections die while sending something
> to the client. There are another about 0.2% where the client didn't properly
> close the connection with QUIT, but they happen while waiting for the next
> command.

Here's some data from a small ISP in Moscow, Russia (the log is for
about a day):

cannabis!root:~# grep -c 'Authentication passed' /var/log/popper
28042
cannabis!root:~# grep -c 'Authentication failed' /var/log/popper
699
cannabis!root:~# grep -c 'Session crashed' /var/log/popper
198
cannabis!root:~# grep -c 'Connection timed out' /var/log/popper
46
cannabis!root:~# grep -c 'Another MUA active' /var/log/popper
7

About 0.7% of the sessions crashed while sending something to the
client, and almost 1% didn't terminate with a QUIT.  There may also
be a few cases where people don't interrupt a slow connection (to
re-dial or re-connect over a network) because they know they wouldn't
be able to continue the transfer of a large message.

I'm not sure if the problem is worth an extension to the protocol,
but as long as this is an optional protocol feature and there're not
going to be any clients that will refuse to support older servers, I
don't see any problem with defining the extension.

Signed,
Solar Designer