D-Link Forums

The Graveyard - Products No Longer Supported => Hubs and Switches => DGS-1216T => Topic started by: TimotiSt on February 06, 2010, 01:57:21 PM

Title: STP problem with linux bridge
Post by: TimotiSt on February 06, 2010, 01:57:21 PM

I have used a revision A DGS-1216T for years with no problems in an STP setup with HP, SMC and 3COM switches.

I bought two revision D switches so i can dual-home my linux production servers with STP. The switches are interconnected on port 16, the server's NIC1 goes to switch1, NIC2 to switch2. Port 15 serves as uplinks to a Juniper router cluster. I have the exact same setup up and running with Blade Networks switches in an HP blade chassis.

Now, when I initially connect the two D-Link switches, the spanning tree builds fine, switch2 marks it's port 16 as the root port. When all servers are connected only to switch1, everything works fine. If I connect even one server to switch2, both ports become forwarding and I see a packet flood, most likely because of a loop somewhere, and the network segment goes down.

The server ports are untagged VLAN11 or VLAN14, port15-16 is tagged 11 and 14. In a Juniper JSRP cluster, the backup port acts as a blackhole, so that's not the source of the loop. STP enable on all ports, except port 15.

Kernel version: linux-image-2.6.26-1-amd64
Switch firmware: 4.21.01

Any ideas are appreciated!

Cheers,
Tamas

Title: Re: STP problem with linux bridge
Post by: Fatman on February 08, 2010, 08:34:55 AM

Now would be a good time to start discussing STP versions and priorities.

The biggest question I have though is this...  If you are using these switches as redundant interconnects to your servers, why do they have a single uplink.  Or if they have 2 why would you allow the priorities to define a root path as through the other.

Title: Re: STP problem with linux bridge
Post by: TimotiSt on February 08, 2010, 09:54:55 AM

I'm using classic STP. The network setup is as follows:
http://img130.imageshack.us/img130/961/serverstp.png

Bridge priorities:
switch1: 4096
switch2: 5120
linux servers: 65535

Port priorities are 128 everywhere.
It's all-gigabit, so port-costs are 4 (i'm using the 802.1D-1998 short standard).

In the meantime, i came across some linux forums saying that the bnx2 driver drops BPDUs, which very well may be the problem.

Cheers,
Tamas

Title: Re: STP problem with linux bridge
Post by: Fatman on February 08, 2010, 11:24:09 AM

That was my misunderstanding then, I feel like an id10t now.

I hate to take the easy out, but I would look into that driver issue right now.  Is there a workaround or fork that is unaffected?