Multiple interfaces and Spanning Tree

[Ziggy Filek]

I have a cluster of 3 ES80s on Fibre Channel and running SCS ofer Ethernet LAN. I have multiple NICs on each machine, but only one NIC on each node is currently connected to a network switch (using a VLAN). I intended to connect second interface on each node to the same VLAN for redundancy assuming a Spanning Tree Protocol would be ran disabling one of the NICs at any given time. I intended to run IPsafe on it to fail-over the IP address.
Somebody told me he is "almost 100% sure" VMS on Alpha does NOT support the spanning tree algorithm. I find it mind-boggling! Is it true?

I'd appreciate any help. Specifically, if this is true, can I have a fully redundant LAN interconnect i.e connect the 3 WE0s to one VLAN and the 3 WE1s to another, separate VLAN and hope the cluster does not fail if one of the VLANs fail?

THanks

Ziggy

[David B Sneddon]

Why not just get another hub/switch and connect
the second NICs to that device -- totally separate
and should keep the cluster going if you have any
VLAN problems.
Don't configure any IP or DECnet on these, use
them only for SCS traffic.

Dave

[Ziggy Filek]

Dave: The Spanning Tree Protocol prevents loops on the LAN. It does not matter whether you connect the second NIC to another switch: as long as both NICs are in the same broadcast domain (as they must be for SCS to work), both switches must be connected, therefore forming a loop with your twin connection. That's why I contemplated two parallel VLAN's on the same switch (which by the way is fully equivalent to having two totally separate LANs implemented using 2 totally separate sets of switches).

Ziggy

[David B Sneddon]

I personally don't like VLANs. Having two separate
devices also removes a single point of failure.

Dave

[Richard Brodie_1 · Accepted solution]

Somebody told me he is "almost 100% sure" VMS on Alpha does NOT support the spanning tree algorithm. I find it mind-boggling! Is it true?

Yes. Spanning tree is part of the 802.1 standard for bridges. Naturally, a normal host doesn't support it; why should it?

[Ziggy Filek]

Richard: Yes, if it does not pass frames between interfaces, then it's not a bridge.
Maybe I should rephrase the question: Can multiple active NICs on an Alpha be connected to the same LAN for the purpose of running redundant SCS communications with other nodes in the cluster?
The IP issue is irrelevant here, since FailsafeIP can handle it.

[Richard Brodie_1]

"Can multiple active NICs on an Alpha be connected to the same LAN for the purpose of running redundant SCS communications with other nodes in the cluster?"

OK, the answer to that question is that you can pretty much make any arbitrary topology, as long as all the nodes share a LAN somewhere, pairwise. If you have a spare dual NIC you could even make point-to-point connections as a backup. Otherwise, connecting to separate switches and tying the switches together works well, unless something DoSses your LAN.

[Jiri Zelenka]

>>> Can multiple active NICs on an Alpha be connected to the same LAN for the purpose of running redundant SCS communications with other nodes in the cluster?

Yes - and by default SCS use all NICs to form a redundant/shared connection between cluster nodes. All NICs with same performance (SCS rating based on speed, delay, reliability) then form a virtual circuit(s) to other node(s) in cluster.

You can see formed channels (NIC to NIC) via 'mc scacp sho chann' and virtual circuit (bundles of NICs) via 'mc scacp show vc' - depends on VMS version.

Jiri