[1866] in linux-net channel archive
Re: TCP problems in 1.3.59
daemon@ATHENA.MIT.EDU (lilo)
Tue Feb 6 12:23:53 1996
Date: Tue, 6 Feb 1996 09:16:15 -0600 (CST)
From: lilo <TaRDiS@mail.utexas.edu>
To: Linux Net Mailing List <linux-net@vger.rutgers.edu>
On Tue, 6 Feb 1996, Ulrich Windl wrote:
> On 6 Feb 96 at 13:14, Jon Tombs wrote:
> >
> > This is a fact on the internet. From where I am, ftp to just about any
> > site is ~15 hops 1s delay (actually it was last year, now somewhat better).
> > My bash function for ftp does a traceroute in background every 5 minutes,
> > without that I was always getting timedout during transfer.
> >
> > Well the point of this post. I needed the tracroutes before I got a linux box
> > on my desk. This "net path needs a traceroute to wake up" is not a linux
> > thing, but effects all machines. It seems it is the routers in the path that
> > start backloging. If you watch the ftp and traceoute you will see that
> > generally the ftp restarts when traceroute reaches a certain host. I believe
> > overloaded routers have a backoff function for failed routes that gets reset
> > by the action of traceroute.
> >
>
> Would it be wrong then to say "a few essential routers on the
> Internet are broken". Either they implement an agreed upon protocol,
> i.e. IP, or they don't. On the other hand, if it's that way,
> shouldn't a similar code be integrated into the Kernel's TCP/IP code
> (re-awake dead connections)?
Now, there's an idea. Some one of us will have to write and publish a
patch, though.... ;)
lilo
