Reliability & downtime, was Re: So why is "IPv4 with longer addresses" a problem anyway?

[Shane Kerr]

On Tue, 2010-06-01 at 10:25 -0700, Ted Mittelstaedt wrote:
> 
> On 6/1/2010 9:29 AM, Nick Hilliard wrote:
> > On 01/06/2010 14:56, Benedikt Stockebrand wrote:
> >> destination.  NUD ensures that a dead router is detected at an average
> >> 30s, plusminus 15s to avoid synchronization effects.  RFC 4861 has all
> >> the details.
> >
> > It kills me to add further noise to this thread, but this is probably one
> > of the base sources of contention.  It seems that you're happy with 15-45
> > seconds as an operationally / commercially acceptable time period for loss
> > of service in the event of gateway failure / failover.  If you accept this
> > level of service unavailability, then RA / NUD is quite adequate.
> >
> > I don't accept that this is an acceptable operational / commercial
> > proposition for my customers.  Therefore RA / NUD is not viable on the
> > networks which I deal with.
> >
> 
> I don't accept that routers routinely dying is an acceptable operational 
> / commercial proposition for our customers.
> 
> Just for grins I just checked and the last SCHEDULED reboot of our own 
> largest and most complex BGP-running router was 50 weeks ago.
> 
> The last UNSCHEDULED reboot was something like 5 YEARS ago.

Okay, that's one router... multiply that by 10 and you've got a router
doing an unscheduled reboot every 6 months. See how that works? :)

I think Nick was quite correct in identifying that for some operational
environments (like yours apparently) RA is acceptable, but for others it
doesn't fit the needs.

And Ted, please try to stay nice! You are almost calling some of the
other people on the list incompetent. That may be true, but I don't
think you have enough information to know. ;)

--
Shane