what happens to Unix when there are network loops.

[doug mielke]

Silly question follows:
Three times over the past months our network has become unavailable for a few minutes (up to an hour or so). This has caused NFS and CIFS failures on HP/UX, caused Oracle IPC functions to fail, and each time has resulted in a panic to my NAS filer.

Each time, our network group has 'found' a loop in the network, either 2 switch ports (level 3) connected by a single cable, or through hubs. (our users often create mini labs in their cube areas, and have been blamed for these events)

I understand bridging or switch loops can cause failures on those segments, and that spanning tree activity will increase trying to correct or tolorate the loop, but what if I'm not on the affected physical segments? Since I'm a not on those segments, any ill effects should be limited to broadcasts? Even a broadcast storm shouldn't cause a panic, right?

My question is, what is causing failures on my segment when a loop happens on another segment? Or is the Network crew making excuses for some other problem?

[Donny Jekels]

ask your network gurus to look for jabber on your backend subnet (the subent where your servers reside). it sounds like a faulty nic on your subnet.

faulty NIC's can come and go as if nothing happened. analyse your syslog.log entries on all your hosts.

good luck.

[doug mielke]

I should add that the description above applies only to our Unix servers' segment.

The rest of the campus goes haywire, with almost all servers unavailable to users (mail, web, etc)

[Sergejs Svitnevs]

The purpose of Spanning Tree is to avoid and eliminate loops in the network. If a hub had been connected to two ports on the switch, then the switch executes the Spanning Tree Protocol and the Spanning tree will automatically rebuild the topology within the confines of the available bridged LAN components and no user data is passing via the LAN during this time.
Ask your network administrator to change the default settings for the Spanning Tree Protocol (example: STP portfast BPDU guard feature would prevent such a situation by disabling the port as soon as STP BPDU is received on the port).

Regards,
Sergejs