Switch loop?

[Mathias Sundman]

I have a network consisting of about 30 Procurve switches, 2524, 4000, 2626, 2824 and 3400.

Lately I'be been experiencing problems with the network, so in order to continue my troubleshoting I'd need to know the following.

1) Is connecting 2 switches, say 2524's, that only has a single CAM table for all VLAN's, with 2 cables, where each port in configured to transport a seperate VLAN untagged an illegal configuration?

Of cource I must configure them so each VLAN is not causing a loop, but I wonder if this still causes a loop, and if so, why?

I'm not using STP as I'm not intensionally trying to build a reduntant network with "loops".

2) Almost same as 1), but I wonder if connecting three switches as a triangle is an illegal configuration without using STP? Still of cource configured not to loop any VLAN.

[Les Ligetfalvy]

Hmmm... Sounds like a recipe for a quantum packet accelerator (AKA blinkenlights). Do you have a drawing of the layout?

[Mathias Sundman]

Hope not ;-)

Attached is a simple example drawing of scenario 1.

2 HP2524 switches are connected with two cables, from port 1 on switch A to port 1 on switch B, and from port 2 on switch A to port 2 on switch B.

Port 1 on both switches are configured as untagged for VLAN 2.

Port 2 on both switches are configured as untagged for VLAN 3.

Port 5 on both switches are configured as untagged for VLAN 2 and connects to some devices on LAN A.

Port 6 on both switches are configured as untagged for VLAN 3 and connects to some devices on LAN B.

The default VLAN is set to "No" for all ports that are in use.

As neigher VLAN 2 or VLAN 3 is configured so they create a loop, I can't see how the traffic on these VLANs could cause any problems.

I do know that I can't connect LAN A and LAN B together on either side, when using 2524, as these can't handle the same MAC address on muliple VLANs, but if these were 2626s which as a seperate CAM table for each VLAN, that should be okay too, right, as long as don't connect LAN A and LAN B together on both side, as that will cause a loop.

What I'm a little afraid of is if the switch itself is generating any traffic outside my VLAN configuratation, like CDP, LACP or some other protocol, that will be looped even though my VLANs are setup not to cause any loops.

//Mathias

[Olaf Borowski]

Mathias,

I would recommend trunking (port aggregation) the two ports on both switches and tagging those ports for all the VLANs you need. This would avoid the problem in any case. The two ports are now considered "one logical" interface with higher bandwidth and redundancy.

Olaf

[Mathias Sundman]

Yes, I know that is the prefered and recommended way to solve this situation. But I need to know WHY using two normal untagged ports would cause problems, as the real network I'm trying to troubleshot consists of over 30 devices inter-connected in a lot of various ways. Some pair of nodes are connected with 2 or more cables as in my example, and some nodes are connected as a triangle, or a circle with more devices.

Understanding exacly how and why this simple example scenario would cause problems will help me to understand why the network behaves as it does.

Thanks for the tip, but I'm not looking for a solution on how to connect these two switches, I just want to know if this really is an illegal configuration and why.

[Mathias Sundman]

Just for the archive:

I've contacted the HP support and they confirmed that the configuration I described is a valid setup and shouldn't cause any problems as long as no device with a single mac address is connected to both VLANs.

The recommended setup is still to use a tagged trunk though to avoid duplicate mac problems.

[André Beck · Accepted solution]

Hi Mathias,

just some additions:

> Is connecting 2 switches, say 2524's, that
> only has a single CAM table for all VLAN's,
> with 2 cables, where each port in configured
> to transport a seperate VLAN untagged an
> illegal configuration?

Normally it is legal, as this is the way to connect multiple VLANs without tagging. Tagging was invented so multiple links become unnecessary, but per se they are perfectly Ok. The actual problem in such cases might be STP.

> Of cource I must configure them so each
> VLAN is not causing a loop, but I wonder
> if this still causes a loop, and if so,
> why?

When using single instance STP (like most of the ProCurves do) there is a chance that for STP, there is still a loop, even if it isn't there for the actual VLAN, because STP sees a topology merged from all VLANs. This can only be solved with multiple instance STP (ideally PVST, to some extent also MST, but MST is painful when it comes to interworking).

> I'm not using STP as I'm not intensionally
> trying to build a reduntant network with
> "loops".

This would get rid of the abovementioned problem, but is a dangerous design in itself. Any reconfiguration that actually creates a loop will blow your network. I know this should not happen, but in this case it is just a matter of mixing up port numbers or VLANs when changing VLAN membership and if you ask me, yes, it will happen sooner or later. There will be a day when somebody (or worse, some automated software) will change the wrong port to the wrong VLAN and kaboom.

To play it safe (or better said somewhat safer, as STP software can still fail on switching hardware that continues forwarding frames, causing network meltdown anyway) it is recommended to activate STP (probably RSTP) on all gear that has any chance of beeing looped, using tagging and when necessary port aggregation for all inter switch links. With single STP you will likely have to put every VLAN on every ISL so that spanning tree and physical topologies (which are identical in this case) match with every single VLAN topology.

[Les Ligetfalvy]

We never did see a drawing nor did we see any explanation of scenario 2.

>2) Almost same as 1), but I wonder if connecting three switches as a triangle is an illegal configuration without using STP? Still of cource configured not to loop any VLAN

[Mathias Sundman]

Les,

I was satified with getting an answer to scenario 1 as I believe it's the same thing. Scenario 1 is just the smallest type of ring configuration you can do, two cables between two switches. With sce 2 I ment tree switches connected as a triangle.

Attached is a drawing of what I ment.

//Mathias