[sclug] the case of the exploding switch

[alan c]

Tom Carbert-Allen wrote:
> Ok so I volunteer to look after a local school network. They have been 
> happy with there Linux box server and gateway for the last 5 years 
> without issue (using a recycled p3 500) but are now having some very 
> weird network problems.
> 
> First sign of problem was 4 months ago I arrive on site to find the 
> central switch is 100% dead. I figure after 5 years in a hot cupboard it 
> just gave up (was only cheapy ebuyer special). I replace it with a US 
> robotics unit I was given and walk away expecting at least another 5 
> years of trouble free packet forwarding. One month later I get a call 
> again and find this switch is now 100% dead too. I then test every 
> connection to the switch end 2 end with a cable tester and they all show 
> ok. I connect switch number 3 and then test each socket with my laptop 
> and they all give me 100mb full duplex and no packet loss. I run out of 
> time so leave again very confused.
> 
> Today I get a call again and find the switch is not dead but all the 
> lights are on (even for sockets with nothing plugged in) and no packets 
> are forwarding. I come armed with a HP Procurve this time. I plug the 
> procurve in and it lasts 3 minutes before shutting off, with nothing in 
> the logs, it just freezes, I reboot it with nothing pluged in and it's 
> failing to come up with error about transceiver VRM failed primary test, 
> I call HP support for a RMA and they give me one but say I must have a 
> shorted cable somewhere or a faulty ethernet device. I plug in another 
> ?20 ebuyer special cheap and cheerful switch and it works fine for 
> today, but for how long.
> 
> Can anyone suggest to me how I might go about finding which cable/piece 
> of equipment is faulty and causing the switch's to all die? I have a 
> basic cable tester but that just test for continuity and pinout 
> correctness. All the kit is connecting ok and not showing lost packets.
> 
> Thanks in advance for your help.
> 
> The only idea I have so far is to buy lots of smaller 4 port switch's 
> and use them instead in a switchy mess of interconnects. That way I can 
> atleast narrow it down to 3 cables/devices.

If it is a cable (mechanical) issue then cable testing will hopefully 
include physical movement of the cables - both ends, with an awareness 
of which cables are most likely to get moved in normal use  also.

The power for the switch is also worth considering carefully. The 
mains supply could be very spikey at times maybe. I had a front door 
bell which was actually a buzzer of the old fashioned moving armature 
type. Nearby there was an automatic daylight sensing lamp. the lamp 
kept failing. I am sure this was because the inductive spikes that the 
buzzer put onto the mains in that vicinity was the cause. Is the mains 
used or abused by unusual laboratory use locally I wonder?

In this case obviously a UPS etc would good to try.
-- 
alan cocks
Kubuntu user#10391