Offtopic (sorry) - Network Switches losing the plot.

Offtopic (sorry) - Network Switches losing the plot.

Post by David Pree » Tue, 23 May 2000 04:00:00



I know this is off topic for the group, but I like this group and
don't really feel like trawling around to find another one. The last
thing I need is another subscription, and someone here will know this.

The tail end of my day was decimated by an Intel 24 port switch that
had by FreeBSD box (running several X apps) connected to it. Said
switch has one (maybe two) other 24 port Cisco switches hooked into
port #24. By around 3PM we were starting to suffer occasional outages
of a couple of seconds each and by around 6 were down to almost zero
uptime. A trip to the server room shows said switch with all its'
green link lights off, then a few recover and finally all of them
within about a half second - leading to blackness again for another
2-3 seconds.

Struggle struggle struggle, plop........... struggle struggle
struggle, plop.

Clearly, it had lost the plot. Our network admin (an NT head) swore
blind it was fine until I found something blunt to beat him with.

But I've never seen this before. What was causing the switch to lose
it? From an uneducated point of view it seems that building the list
of MAC addresses connected to port 24 caused the switch to function
correctly in the short term, then buffer overrun and reboot itself.
Repeatedly. What were we doing differently in the afternoon? OK, you
weren't there - silly comment.

I guess what I'm asking is: When switches fail - why? And for gods'
sake, how can I prevent it?

Dave, who hates being stopped while in the zone.

----------------------------------------
David Preece   http://www.dmpreece.net/
Software Engineer,        Wellington NZ.

 
 
 

Offtopic (sorry) - Network Switches losing the plot.

Post by David Malo » Thu, 01 Jun 2000 04:00:00



>I guess what I'm asking is: When switches fail - why? And for gods'
>sake, how can I prevent it?

We've seen some switches die if they see to many addresses, so
that's certainly one possibility.

        David.

 
 
 

Offtopic (sorry) - Network Switches losing the plot.

Post by Torfinn Ingolfse » Thu, 01 Jun 2000 04:00:00




> >I guess what I'm asking is: When switches fail - why? And for gods'
> >sake, how can I prevent it?

> We've seen some switches die if they see to many addresses, so
> that's certainly one possibility.

Yes, most switches have a limit on how many adresses they can hold in
their MAC table(s). Normally, when the MAC table gets full, the switch
just flushes it and starts loading it from fresh again, no problem.
The problem starts when the switch a) uses all CPU to reload the MAC
table (thus "starving" other necessary processes in the switch),
or b) is unable to "hold on to" the contents of its MAC table for a
reasonable length of time (seconds? minutes?), it then becomes necessary
for the switch to "flood" all ports with traffic.

Each vendor (Cisco, Intel, Nortel, networks, etc.) have their own
"strategy" on how to deal with MAC table problems (I have experience
with Cisco, Nortel Networks and a couple of others. Not Intel, though).
Generally I find that Cisco switches are the ones that can stand the
hardest "beating" without losing it. Beware of large unsegmented LANs;
no switch is happy if it has to talk to about 1000+ devices on one
network...
--
Torfinn Ingolfsen,
Norway

 
 
 

1. Offtopic (sorry) but I need yer expertise

Hey gang,

I'm posting this here cuz I just know someone in one of these newsgroups
has the answers I'm looking for.

Got a huge old (1991 or so) server from a friend the other day. He
didn't even know if it worked or not (seems someone left it in his attic
before he moved in), he *thinks* it runs unix or OS/2, but is not sure.

Here is what I am sure about:

1:    It's huge, dual (*maybe* quad) processors covered by enormous
heatsinks, from the size of the chips, I'd say some type of 486.
2:    All intel chip names everywhere inside
3:    8 EISA slots and one extra slot that looks like a microchannel
slot (but may be some type of extra memory slot)
4:    3.5"/5.25" combo drive, 250 mb tape drive, cd-rom (unknown speed),
213 maxtor HD (isa, not scsi)
5:    4 slots of 72 pin simm ram (8 megs total on it)
6:    on the back of the machine it says "intel" and "produce code:
XBASE8TE8F"
7:    damned irritating BIOS (all 4 pages worth)

I can't get it to boot up, because *something* in the BIOS is incorrect
and halts it (BIOS says something is wrong, but doesn't say what), so I
can't get much further than some startup code.

8:    front panel is lcd screen that gives the error codes (702 is only
one I remember offhand) and something about the countdown timer and bios
stamp (I think), but no clue to the error code meanings.

I'd like to boot this thing and find out what it is (obviously) and
maybe put some OS on that recognizes multiple processors (any
suggestions guys?), but I need some more info. HELP is mucho appreciated
:)

P.S. onboard 1 meg ISA Adaptec IDE controller (if that helps)

2. samba printer sharing problem

3. Help.Sorry offtopic.

4. VPN to Linux server behind NAT router from XP Home?

5. Wanted::Scientific Plotting program (with error bar plotting)??

6. Can't see DOS partition... Help!

7. Plotting package as good as Sigma Plot or Kaleidagraph

8. XFree86 : Curved windows in hi-res?

9. ANNOUNCE: FMRTG: File based Web plotting tool with Solaris performance plots

10. Need Help sending UNIX plots to a NetWare plot queue

11. Wanted::Scientific Plotting program (with error bar plotting)??

12. Linux Network down work with 1000 MBit Switch but with old 100 MBit Switch

13. Sorry I lost the thread but ....