(#003) - The Myth of Auto-Negotiation
by: Jim Huddle CNE CNE5 CBS MCSE ES-RC, send
watch Dennis Miller's show on late night HBO? Mr. Miller usually
has a topic he begins joking about. After a few jokes he generally
says something like " I don't mean to go off on a rant..."
and then does just that for several minutes. The rant is often
humorous, as is his intention, but is also rather biting about
the serious point of the topic. At the end he says, "but
that's just my opinion, I could be wrong."
don't have any jokes, but I do have a rant, and it is just
my opinion. Like as not I'm wrong, but I can only go on experience.
My current particular rant is about a part of Ethernet switching
switching was introduced to help alleviate the problem of
collisions. Traditional shared Ethernet networks operate by
allowing one device to pass one data packet on the wire at
a time. The devices, whether they are servers, PCs, print
servers, etc., all have to negotiate for time to send their
packets of data across the wire. When the Network Interface
Card (NIC) in a device senses that no packets are being transmitted,
it will begin sending its own packet. If another device also
begins sending at the same time there will be a collision
on the wire and both devices will back off for a random amount
of time before trying again. On a small network, this usually
passes unnoticed. On a large or very busy network, the resulting
time loss can be a significant problem.
the communications between devices attached to a switching
hub are privately negotiated there are no collisions. This
was a nice improvement. A couple of other advancements came
along: duplexing and 100 megabit (Mb) Ethernet. Devices with
compatible NICs could send and receive simultaneously. In
theory doubled the speed of network communications. 100Mb
Ethernet boosted the speed of traditional Ethernet tenfold.
this is great, except for one problem. Networks ended up being
built with a mixture of speed and bandwidth capabilities.
Vendors of networking equipment had to come up with a switch
that would accommodate the differing speeds and duplexing
capabilities. What they came up with was Auto-negotiation.
The idea was slick. Devices supporting auto-negotiation would
negotiate the highest speed and determine the duplexing mode.
So you could have 10Mb and 100Mb devices attached to the same
switch and not have to worry about setting the individual
ports to match the capabilities of the device being attached.
Just plug and play. Ah, I'll bet that last phrase caught your
used switching equipment from many manufacturers, and more
kinds of NICs than I'd rather think about. The one thing in
common they all seem to have is that auto-negotiation, uh,
probably been there. You get in a shiny new switch or switch-router.
You may be using it to speed up the engineering department's
CAD downloads or maybe it's an upgrade to your core. You are
optimistic. After all, you didn't buy Joe's NoName Switch-Router,
you bought a major manufacturer's best offering. So you attach
a ton of hardware, upload the configuration file that you've
meticulously created, bring the unit up and smile confidently
as the box goes through it's self check and boot process.
All the pretty lights turn green, the switch-router is up,
and just as you're congratulating yourself on how bright you
are one or more, probably more, of the ports start blinking
red or just shuts down altogether.
slightly. Check the box through the laptop you've judiciously
attached to the switch-router's console port. Everything looks
OK. Config is right, ports don't show any errors, and the
box appears fat, dumb and happy. You do the stupid stuff like
checking the cables and fiber links. You check the servers
and routers just to be sure they're still running. Everything
appears to be operating, but the ports are still dead or constantly
initializing. So you recheck everything again, even applying
the NOS and firmware patches you've been meaning to get around
to. Finally you reboot the new box. It comes up again but
this time all the ports stay up. You shrug, concluding it
was just a warning from the Cosmic Muffin that you've been
lax in sacrificing users.
you pack up the laptop and begin thinking about the next project.
A short time later you start getting calls from users complaining
that they can't access the servers or the Internet. Going
back to the box, you find several ports have decided to become
blinking Christmas displays. You do the trouble shooting again
and still all looks fine. Re-attaching the laptop, you scan
the down ports. They still show fine, just down. Next your
boss comes in wanting to know what you are doing about this
problem. He or she has just got a call from the Marketing
department and they are in a dither because they can't access
that new Internet site that promises to teach them how to
lie more sincerely. You're getting desperate. Others are panicking.
select the port that connects to the Internet router, open
its configuration screen and look at it's setup. You recall
that about six months ago you had a problem with one user's
PC that wouldn't connect until you turned off auto-negotiation
on their NIC. You figure it's a long shot but worth a try.
So you change the ports configuration and turn off auto-negotiation
and manually set the connection speed and duplexing. Suddenly
the holiday season is over and the port starts working showing
only the reassuring green lights. You go to each dead port
and manually set them and they all come up.
watch the ports for a time, only half listening to the nattering
of your bosses about how "This kind of thing just can't
be allowed to happen" and "We need to have a meeting
(or ten) to decide how to prevent the lack of meetings."
Meanwhile you're studying the packet statistics and idling
thinking that the overdue sacrifice is a toss up between the
Marketing department and your own management. Either would
offer significant enhancements to your sense of well being,
but you worry the Muffin might get so serious a case of indigestion
that it would curse you by placing a major whiner next to
you have an inner debate on whether or not it would be worth
it, one of the ports attached to a server goes down again.
Your bosses begin wondering if maybe you shouldn't put the
old equipment back online because there was never any trouble
with it. Never mind that it was so old that the vendor no
longer supported it and the support folks actually sniggered
at you whenever you called about getting replacement parts.
You go to the server and pull up it's network configuration.
Sure enough, the NIC is set to auto-negotiate the link. You
turn it off and set the connection's speed and duplexing manually.
The link comes back up. Cursing your lot, you decide to check
every device coming into the new switch-router and manually
set the NIC as a precaution. You miss a switch that feeds
one area, but you don't notice it as it's continuing to pass
a couple of months after that the network goes about its business
of moving packets. The folks in Marketing become more proficient
prevaricators and for now they are happy. The bosses mercifully
go on extended retreats after cautioning you several times
not to fix any problems while they are gone, and you've moved
on the other projects that need to be finished.
day you get a call from a user that uses the switch you neglected
to set. She can't login. You check the core switch and sure
enough that port has gone south. You check the user's switch
and discover that it is still set to auto-negotiate the link
to the core. After turning that off and setting the link manually,
the link is restored. You wonder why it suddenly stopped working.
suppose the bottom line is this: You can't trust auto-negotiation
to work. Sometimes it works flawlessly for months and years
then suddenly fails. It's become the first thing I check when
I have any connectivity problems. Any new switches added to
the core will now have auto-negotiation on the core link turned
off by default. Auto-negotiation is a great idea and maybe
one day it will work consistently, but not today. My advice
is to turn it off and forget it's even a feature to consider
using. But, like Mr. Miller says, that's just my opinion,
I could be wrong.
to the Editor are welcome and occasionally abused in public.
Send e-mail to: email@example.com