Re: how to deal with hosts that are down

Janos Farkas (Janos.Farkas-nouce/priv-#.c53ZCnhNSqTNOtnvYuHvDb394y@lk9qw.mail.eon.ml.org)
Thu, 24 Sep 1998 12:40:39 +0200


[I hope you don't mind being Cc'd, linux-kernel is soo slow...]

On 1998-09-23 at 16:56:52, Stefan Monnier wrote:
> .... What I have in mind is an OS-wide solution, so that if a host is
> down and someone tries to access it via resolv/ftp/telnet/http/younameit they
> will immediately get a timeout (since the host is know to be down because of an
> earlier timeout, why wait 2 minutes before returning the timeout to the
> application ?). I'm quite wawre that I haven't thought this out in detail
> and that it isn't quite so easy to solve, but I'm fairly sure something
> can be done.

In a sense, it's already done, so you need to be more specific what you
want and prove that it works in other cases too.. :)

2.1 currently "knows" about hosts on the local network, via its
generalized neighbour discovery, of which ARP is just a special case.
When you are on IP 10.20.30.1, and 10.20.30.11 is not on the network, a
telnet to that host will timeout with "No route to host" in about 4
seconds. However, when that host WAS reachable a few minutes ago, and
still is in the ARP cache, the failing telnet is taking much, much
longer.

However, this is still good, IMHO. You don't want all you network
connections suddenly be dropped, since you are reorganizing the network
cabling, and unplugged your network for a second.. When connections are
open, they too take a few minutes to timeout, and this is even required
by RFC's, AFAIR, for similar reasons. (At least there is a real good
reason to ignore sporadic "no route to host/network" messages on the
network level.

Even when you *know* that on ethernet, nothing will be delayed by two
seconds, you can't be sure someone else don't need it *still* working
after two minutes of cable arrangement.

-- 
Janos - Don't worry, my address is real.  I'm just bored of spam.

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/