Switches, Hubs, and Modems
Showing results for 
Search instead for 
Did you mean: 

RSTP versus Meshing and LACP

Gerard Lubberink
Occasional Visitor

RSTP versus Meshing and LACP

At a customer site we build a redundant network using two 5308 switches in the core and several 2600 switches at the edge. The two 5308 switches are trunked with a four port trunk, connectivity to the edge switches is achieved using multiple gigabit links. The multiple links provide redundancy using RSTP. The servers are connected with two links connected to both of the 5308 switches.

A HP engineer has visited the customer site and advised to:
- block spanning tree on all edge ports
- use no lacp command on all interfaces in the network
- use meshing as spanning tree because it is a better implementation of the spanning tree protocol

In my humble opinion the last two point are not correct blocking lacp also blocks 802.3ad which is used by HP teaming and the trunk and meshing removes the layer 3 functionality.

What do you think?

Shiraz Malik

Re: RSTP versus Meshing and LACP

it depends how u've configured the trunk. If you have done it statically, then its no harm in blocking LACP on all ports. but if u r using LACP, then obviously u can not disable it on the ports where u r using trunks. but still u can disable it on rest of the ports. Meshing is obviously a better way to configure redundancy and to eliminate loops as it does not block any port and aggregates the bandwidth. I dont see any issue with these recommendations.

Occasional Advisor

Re: RSTP versus Meshing and LACP

Hello Gerard,

Disabling dynamic LACP on interfaces where you do not need it, will improve link performance in that the port will come up faster. Similar to what Malik is saying.

HP Meshing and IP routing indeed do not operate together. If you need routing over layer 3, you cannot use HP Meshing.

An additional factor you may want to consider is that the 2600 switches do not support meshing. If I understand your topology correctly, you will be able to run meshing only on the core 5300XLs and will still need to run spanning tree on the 2600s to get the loop out of the redundant links between the 2600 and meshed 5300XLs.

Furthermore, you will already make full use of the bandwidth between the meshable devices by using spanning tree. You have a 4-port trunk running. I assume the spanning tree root priority and costs are configured in such a way, that you will not have your 4-port trunk blocked, but that the blocking port is somewhere on the 2600s.

In short, based on what you say here, I would not use meshing. It doesn't add anything to what you already have and you'd have to get rid of routing.

André Beck
Honored Contributor

Re: RSTP versus Meshing and LACP

Well, as always, the only correct answer is "it depends". But I'm with you:

Switching off LACP gave me a lot of headaches lately, ISL ports didn't find each other after giving them a "no lacp" on both ends. This was both 3400cl vs. 2600 aand 3400cl vs. 3400cl. Having passive LACP on all access ports is basically a good idea as it allows for transparent support of LACP capable end systems (like you said HP and other vendors NIC teaming). It does hurt a bit by giving you additional 2 seconds for a link up, but given the rest is RSTP, you can normally live with that. ProCurves cannot block (R)STP at all, so what is probably meant is to configure all access ports as edge (or portfast respectively), which is a good idea. It is even more good with RSTP due to the way it works, you *never* want an access port to be misinterpreted as a non-edge port. Then again, RSTP normally finds out about this by itself, just not always fast enough.

Now for Meshing: It is not "a better implementation of the STP". It is no STP at all, except at the edges. Meshing is a completely proprietary "routing at L2" style solution similar to other vendors solutions in this field, compare Cabletron SecureFast for instance. Now the basic thing about these solutions is, except for HP meshing, they are *all* *dead* today. They introduce inter vendor problems to start with, and sooner or later they will be abandoned by the inventing vendor itself (this is no rant about HP, just experience). Just count which of the ProCurve boxes available today do Meshing.

And then the real problem: Meshing contradicts routing. To me, this is a complete No-No. I'm not going to build inflated broadcast domains any longer, the trend is completely different, it is back to localized small broadcast domains and routing between them. It is actually away from campus spanning VLANs. Thus, meshing is no option IMO, but of course others mileage may vary.

If you are still free in your topology and addressing design, rather consider to build a wellformed collapsed core (core and distribution using the 5308xl boxes) and L2 access based on dynamic routing and XRRP, with VLANs not spanning further than from the access switches to the distribution (or core in this case). It is the thing to come, not ever inflating broadcast domains.

Just my 0.02 EUR,