[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Pan-users] Pan 0.13.3.92 download connections
From: |
Duncan |
Subject: |
Re: [Pan-users] Pan 0.13.3.92 download connections |
Date: |
Mon, 10 Feb 2003 13:27:25 -0700 |
User-agent: |
KMail/1.5 |
On Mon 10 Feb 2003 01:28, Brian Morrison posted as excerpted below:
> On Sun, 9 Feb 2003 20:51:38 -0700 in
> address@hidden Duncan <address@hidden>
>
> wrote:
> > Finally, a connection may occasionally die, without properly being
> > terminated. This is especially likely on a bad or noisy analog
> > connection (like dialup), but may happen due to other factors as well.
> > When this happens, the server
> > doesn't realize that connection is dead until its timeout expires.
> > During the period until timeout of the dead connection, you won't be
> > able to make another one to fill that hole, so you will effectively be
> > limited to two connections (or one, if two are dead but not timed out,
> > etc).
>
> Ah, that sounds a lot like the problem I've seen a few times where I get
> a 503 Timeout error when I try to open a new group or read a new
> article. I wonder if in fact they could be related?
It's possible. Even if the number of server connections is set correctly, if
your behavior is to start a task downloading in the background, then go to
another group while it completes, you are more likely to get the error now,
than before, because the d/ling will take as many connections as possible up
to the maximum allowed (provided there are that many individual posts or
parts to be downloaded, it doesn't break up a single part). Now, PAN is
designed to stop the old task and start the new one (with as many threads as
necessary on the new one) if it thinks the connection limit has been reached,
according to Charles, and I see that behavior here, but if one of those
connections died, PAN may time it out before the server does, and try
periodically to start a new connection, creating an error each time, until
the server times out the dead connection as well. Thus, since PAN makes more
efficient use of the connections available now, a dead one is far more likely
to create an error than before, if you weren't deliberately setting up
multiple tasks to use them all back then.
Does that make any sense? <g>
Most servers send a connection refused error rather than timing out, when the
connection count gets to high. However, some may simply drop the new
connection attempt, resulting in the timeout that you see. Of course, the
other possibility is that it really IS a timeout, caused by overall
connection problems, not a dead single connection.
--
Duncan
"They that can give up essential liberty to obtain a little
temporary safety, deserve neither liberty nor safety." --
Benjamin Franklin