prg
> > Could it be that the switches and hosts are coming back up after a
> > power outage in such a manner/sequence that the problem converges
on
> > the suspect location ...
> This is exactly what I'm worried about.
This might be a one time thing related to computer boot ups and switch
learning rates. But it may be useful as a locational clue to a problem
source. Or maybe not :(
Quote:> > ...and craps the spanning tree -- if you're running
> > spanning tree, that is.
> If "spanning tree" is running, I don't know it. I have SuSE 8.1 and
> 9.1, pretty standard setup. Linksys consumer switches (the $50 at
Best
> Buy type) and a WRT54g. These don't provide the spanning tree
protocol,
> do they? (Googling, I can see that Linksys WET54GS5 does do spanning
> tree.)
This was more just a check -- have noted in the past year that a number
of commodity switches are offering spanning tree. Doubt that it will
even engage without another spanning tree switch in the fabric. Guides
don't speack much about configuring it -- just on or off.
Quote:> > I've had times when MACs began appearing on different sides of a
switch
> > for no apparent reason. This can bring the switch(es) to their
knees.
> > Hard to "see" it without sniffing the wire and the problem source
can
> > be difficult to track down.
> No loops in the physical wiring. I'm absolutely sure of this.
> But... what if a host (and its MAC address) are moved from one part
of
> the LAN to another?
> I move a laptop (Apple OS X) all the time from wireless (with a NAT'd
> non-routable IP) to wired (with fixed IP) at the same OR different
part
> of the network to outside the building (i.e. yet another IP somewhere
> else on the internet). Could moving this Mac's MACs be causing the
> problem?
Well, at the switch level, only MACs and ports are used in the lookup
tables. This is strictly level 2. If you are moving within the same
switch group/lan and the ageing timers are long enough, your old
MAC/port will still be present when you hook up elsewhere -- you
"announce" your presence to the switch with the very first frame you
send out.
Quote:> FWIW, the Mac laptop has two nics, one wired and one wireless.
> Different MAC addresses. OS X in most configurations automatically
> changes from one nic to the other as needed. Recently someone opined
in
> alt.os.linux.suse "a host with two nics on the same LAN is trouble."
This is true at the IP level, especially re: arp cache/table of
MACs/IPs as gateways can get confused when they have MAC/IP pairs
moving around. It's similar to MACs moving about in a switched LAN.
Any host/nic setup that produces "gratuitous" arp (announcing arp) can
exacerbate this. But these will (or at least should if not turned
off/filtered) return ICMP errors.
If only one nic is active at a time, no problem. Both at the same time
on a Linux box is usually trouble because Linux stack is designed to
respond for any host IP out any nic (a sort of failover "feature").
Don't think Mac stack behaves this way -- it's BSD based, afaik.
Quote:> I
> wonder if I should start being extremely careful that no laptop ever
> has ethernet plugged in which it's wireless card is operating. Anyone
> have comments on two-nics-same-LAN?
Do you gain anything with both nics up simultaneously? Failover and
bonding are the usual reasons for wanting both up. The route path to
the nics is likely different -- true in your case? Normal Linux
operation does not use metrics and with its quirky responding behavior
it's quite possible that packets come in on one nic but out the other
-- routers and GWs upstream don't like this bouncing MAC address in the
arp traffic and the ethernet frames could, I imagine, make switches
bark in the right circumstances.
Quote:> I've seen the problem three times
> - Twice with the moving laptop outside the building
> - Once after a power failure with the laptop in it's usual place on
> the LAN.
> > Resetting the switches clears the tables and you start clean again.
> > After some interval X, the problem accumulates/reappears and brings
the
> > switches down again.
> Yup, that sounds like what I've seen.
> > Double check, then double check again, that there are no loops in
the
> > physical pathways.
> No loops in physical wiring. Is it conceivable that I created a loop
> involving wireless via the Apple OS X auto-network-config when I
> plugged into ethernet with wireless still running? I'll watch out for
> that.
> > Tried swapping out the "bad" switch with one upstream/downstream or
> > sibbling to see if you can make the problem "follow" the switch.
> A little hard to tell, but I do have one suspect switch which I'll
> watch.
I was sorta hoping that you would find someone complaining about eratic
networking, suspect misbehaving nic, and locate your problem source.
Alas, it's never that easy.
I would keep an eye on the moving laptop. Switches should have a
timeout for table entries with no response for XXX seconds (Ciscos are
300) so you might want to check what the timers are set at and change
them or wait before plugging in at a different switch. We've had
trouble with students at school when moving about the right classes --
like several rooms away but the same switch or sibling.
Checking switch configs and stats can help but is usually time
consuming and tedious. You might want to try a sniffer -- I like
ethereal -- and watch for "unusual" ethernet frames, especially what
appear to be multiple, identical frames. Also keep an eye out for a
high volume of broadcasts or anything that looks like an errant "keep
alive" packet.
good luck,
prg
email above disabled
by