lars.rasmussen at gmail.com
Mon Jul 25 13:00:13 MDT 2005
On 7/25/05, Hans Fugal <hans at fugal.net> wrote:
> We agree that power supplies fail statistically more frequently than
> NICs. You (or someone, I can't remember) think that NICs aren't worth
> making redundant because they never fail - that's what I don't agree
switching contexts? "NICs never fail"? Wasn't said.
The CARP setup I advocated actually gives MORE protection against NIC
& power supply failure because in addition to protecting against NIC
failure on one box, it can protect against other possible failures
(cable failure, cables being unplugged, switch failure) on any of the
nodes in the cluster, including some of the failures you previously
mentioned, while providing a single "virtual" IP gateway to the
internal network. A CARP cluster offers more uptime due to redundancy
and failover handling than the bonding solution you proposed, but is
more complex to implement.
> > > I don't really care how power supplies compare to NICs, the point is
> > > that NICs fail. Even more important, switches fail, switches lose power,
> > > cables between switches and servers and other switches fail, and dumb
> > > system administrators accidentally unplug cables. I'm quite sure all
> > > those problems combined are more frequent than failing power supplies.
> > This last statement is a good example of the 'Hasty generalization' &
> > 'Appeal to authority' logical fallacies.
> I disagree. My original facetious comment was a hasty generalization,
> that's what made it funny. Ok, so you're not laughing. I don't see
> anywhere above or in the thread where I appealed to any authority but my
Precisly. The statement, "I'm quite sure all those problems combined
are more frequent than failing power supplies." cites yourself as the
authority(therfore an appeal to authority) and is a generalization
based on an admitted small sampling of your own experience.
> I admit to not having done a scientific statistical study, but it
> doesn't take a study to recognize that NIC connectivity loss happens.
I don't see where you're going here - am I supposed to adopt a new
premise & context of our arguement that "NICs never fail"?
More information about the PLUG