MST operation on Procurve switches

[Regnar Bang Lyngsø_2]

Hi,

does anyone have experience with MST (Multiple Spanning Tree) on HP Procurve Switches?

I've been running a few tests in my test lab. I am using Procurve 2650 (running firmware H8.53), Procurve 5308XL (running firmware E9.03), and Cisco Catalyst 2950 (running IOS 12.1(22)EA4).

I have 2 issues:

1. Interoperability. Running MST on 2950 and 2650 - all instances mapped to instance 0, same config name, same config rev - the switches do not form a region.

Enable debugging on Catalyst 2950:
Switch#debug spanning-tree mstp sanity_check

18:00:19: STP: Fa0/24 Sanity Check --> BPDU with invalid v3 length(16384)
18:00:21: STP: MST00 port Fa0/24 agreement from non-root/design

Shouldn't both switches use the same IEEE compliant spanning tree version (I have configured MST operation on both boxes)?

The Cisco talks RSTP to the Procurve:
Fa0/24 Root FWD 200000 128.24 P2p Bound(RSTP)


2. Instance mapping. On the Catalyst 2950, I can map VLANs that do not exist on the switch, i.e.:

spanning-tree mst configuration
name RBLTEST
instance 1 vlan 100-199
instance 2 vlan 200-299

On the HP Switch I am only allowed to map VLANs that are configured on the switch:

hp2650(config)# spanning-tree instance 1 vlan 100-199
100-199: No VLAN found in the specified range.
hp2650(config)#

This makes it very difficult to do a proper redundant design with active/active links (same behaviour is observed on the HP5308XL).

If the HP's behaved like the Cisco box I could have one campus-wide region with identical instance mappings, but VLANs only configured where needed.

To have all VLANs defined on all switches in the network is clearly not an option. Is this a bug on the HP switches or is this the intended behaviour?

Regnar

[Regnar Bang Lyngsø_2]

Great! Cisco has a bug in their 802.1s implementation (according to Ethereal - and various links - e.g. http://groups.google.com/groups?selm=cepv6v%24iqg%241%40news2.netvision.net.il ).

So now I'm down to the instance mapping on Procurves.

[Shiraz Malik]

Regnar,

procurve 2650 does not support MST. the only procurve edge switch that supports MST is 5300. Dont know much about cisco.

regards,
shiraz

[Ralph Bean_2]

Hello All -

If by 'MST' you mean 802.1s (Multiple Spanning Trees), the ProCurve 26XX models support it as of H.08.53 per the Release Note at http://tinyurl.com/ayg4l (see page 32).

Regards,
Ralph

[Regnar Bang Lyngsø_2]

Exactly. That was why i did the test on FW 8.53 on the 2650.

With HP not allowing instance mapping for VLANs not present on a switch makes it difficult for me to see the great advantage of 802.1s in a Procurve based network. Any thoughts?

As I see it, one region is the way to go - among other things - redundancy on layer 3 with OSPF should work like a charm. On the other hand - have all VLANs (>200 in our network) on every single access switch is simply not an option.

Guess I'll have to settle with RSTP.

[Frank Benke_1]

So, to share my point of view shortly ...

MSTP (802.1s compliant) found its way into the latest firmware revisions for the ProCurve 26 and 28 series switches and is no longer available not only with 34 and 53 series. I tested it in a fairly large network containing all these 4 series switches and it works fine.

The 2900 Cisco whereas does something proprietary since natively Cisco supports only per VLAN spanning tree (PVST) . I made the interoperability only work when I configured a MST instance per VLAN on the Pro Curve Side and put the IST (Internal Spanning Tree) sure in VLAN 1. Then activated the PVST on the Cisco side and have the management VLAN / VLAN 1 connected to the IST / VLAN 1 on the ProCurve untagged.

The Ciscos and the HPs will not behave the same way (at least not in the near future) and you will need to design a work arround. The one mentioned worked for me.

regards frank

[Frank Benke_1]

but I have only one CISCO 2900 ... to test interoperability on that side :o( so that is the limit in my scenario.

[Regnar Bang Lyngsø_2]

Cisco has 802.1s in later versions of IOS on the 2950 (but at least the version I tested was inconsistent with the examples in the standard when it comes to the packet format) - I am aware of the PVST vs single instance spanning tree - I am running a mixed vendor setup - with about 50/50 Cisco/HP equipment - Cisco being phased out - and spanning tree is working in this setup - MST would be so much better.

I think I will try to make a drawing one day - explaining why I think that 802.1s could give most of the benefits of Ciscos (R)PVST+ - and why I think HPs implementation makes it difficult.

Regnar

[Regnar Bang Lyngsø_2]

to clarify.

Given the topology in the attached file:

1. I would like to have VLANs 2xx and 3xx defined *only* on COR2, COR3 and ACCx switches.

2. I don't want to have my core VLANs (VLAN 1x, VLAN 2x, and VLAN 3x) defined on ACC-switches.

3. I would like to have only *one* region.

4. All switches support 802.1s

The only thing that prevents me from doing it on HP switches is the fact that I can't define instance mappings for VLANs not defined on a switch. E.g., I can't define VLAN 12,13, and 14 to be mapped to instance 1 on switch ACC1.

With Cisco switches I am allowed to do this mapping (I am talking 802.1s aka MST - not Cisco's proprietary (R)PVST+ which has one spanning tree per VLAN).

Regnar

Regnar