[plug] [network newbie] diagnosis sick network

[Craig Ringer]

Sol wrote:
> Hello!
> 
> I'm having an odd problem with a small network. Every so often - every so many days - the connection between the server and the rest 
> of the network ceases. I can't ping the server from the client, nor the clients from the server which is a major bummer considering 
> two clients a thin clients. Yet the clients can ping each other. Whilst I suspect a hardware fault - I usually get things working 
> again by run route and *wobbling* the cable until I get a response and then reset the networking daemon for good measure - but I'm not 
> absolutely certain. It could be a dodgy ethernet card?

Dodgy NIC - easily. Dodgy cable - also very easily. Insane switch - 
distressingly common. Interactions between  PnP BIOS / PCI BIOS and NIC 
- easily. Quirkly linux drivers / kernel build - easily. Interrupt 
handling conflict/issue - easily.

If you're running a patched kernel, now's a good time to stop if you 
can, for debugging. I found that the htb3 patches were bringing down my 
firewall at work... but only about once every 2 weeks.

Replace ALL the cables if you can or at least the server -> hub/switch 
cable. Network cables are bastards and they can cause basically any 
freaky fault, including all sorts of intermittant stuff - don't ask me 
how, some of the stuff I see should be impossible, but they do it. Alas.

Hard-reset the switch or hub, and if you can borrow a spare from a fried 
for a while to see if that fixes it. I have personal experience with a 
network where we beat our heads against the wall for ages - it turned 
out to be the switch.

NICs can be incompatable with switches and each other. At work, any 
Netgear NIC I plug in to the switch (an Acer "SmartSwitch" 24/100, not 
my choice!) just won't work - no link light. Ditto mac G4 machines. All 
work fine through the secondary switch that uplinks to the main one.

Anyway, what I'm trying to say is that basically anything could be 
wrong, but network cables are a good place to start, followed by NICs 
and the switch/hub.

> What I was wondering was if there are any diagnosis tools that I could use to help narrow down the cause? 

Ha! I only wish....
There are, but you can't afford them. *sigh*. A proper cable tester (not 
a connectivity tester) starts at (I think) $1000 + .