Ethernet Probe Order 2006-01-05 - By Tom Sightler
Back
On Thu, 2006-01-05 at 21:13 -0600, Ed Wilts wrote:
> On Thu, Jan 05, 2006 at 06:55:41PM -0800, John Reiser wrote:
> > No, the request should be kept open, because the most sensible way
> > is not there yet today. Utilizing HWADDR takes painful, per-system,
> > custom effort. A default of ascending interface number is simple
> > to understand, simple to implement, and takes care of requried
> > predecessors automatically.
>
> And the first time you have an Ethernet adapter fail, you will hate it.
> Suddenly your system will boot on the wrong networks and all hell could
> break loose. I have a lot of systems with 3 NICs in them and they're
> always predictable. I will gladly take a performance penalty any day of
> the week for a more predictable failure scenario.
Nothing keeps the system from both being sensible in initial allocation
and still using the HWADDR method for locking NIC's to specific
adapters. One does not preclude the other.
> That's what people said about disk devices until one failed. Assuming
> the order of devices could very easily lead to massive disk
> corruption.
In my experience disks fail a LOT more often than NICs, especially good
quality NIC's. In the last five years, on dozens of servers and
hundreds of workstations we've lost exactly one NIC that I can remember
and it still was detectable, just wouldn't actually detect a cable, thus
the order still didn't change.
We've probably averaged 1 HD failure per month in that span (although
they usually come in groups). I'm not sure I would be willing to give
up speed on every server for something where the failure rate is so low.
Later,
Tom
--
Taroon-list mailing list
Taroon-list@(protected)
https://www.redhat.com/mailman/listinfo/taroon-list
|
|
