VSS MEC with stacked 3750X

[josephschung]

Core layer, VSS:

1) 2 x WS-C6504-E

2) 2 x VS-S720-10G-3C

Distribution layer 3, Stackwise :

2 x 3750X

Access layer 2:

4 x 3560x

Inter-switch connection:

1) Each 3750X has one connection to one 6504 and configure the two links into layer 3 MEC with ePAgP.

2) Each 3560X will has two connection to each 3750X and form a layer 2 port channel.

I am not sure if the layer 3 MEC with ePAgP will work across stacked 3750X. Any idea?

Thanks.

[krahmani323 · Accepted solution]

Hello Joseph,

As PagP is not supported accross differents 3750(-X) units of a stack (but LACP yes), ePagp is not supported also. The MEC has to be formed from the same 3750(-X) unit / Standalone 35xx and 37xx do support PAgPand thus ePAgP.

Page 17 and 20 of http://www.cisco.com/web/DK/assets/docs/presentations/VSS_0109.pdf =>

"Since cross-stack EtherChannel does not support PAgP, it can not use enhanced PAgP for dual ACTIVE detection.

Hence the only alternative for cross-stack EtherChannel configuration is to use BFD as a dual ACTIVE detection method"

Hope it helps.

Regards;

Karim

[josephschung]

Thanks Karim for your information. As ePAgP is not helpful in this case, I will user the fast-hello between line card of 6500 to detect the dual-active.

I am working on a three-level design. Thus, I will use layer 3 port channel between distribution and core. I am wondering if I should just use layer 2 port channel and let all eigrp routing happen in the core vss only. What do you think?

Thanks.

[Ashley Georgeson · Accepted solution]

Hi Joseph,

My preference would be Layer 3 down to the Distribution Layer at least, if not down to the Access Layer.

Imagine the size of the broadcast domains and failure domains if the VLANs were terminated on the VSS!  Also they would need to process all your Access Layer broadcast traffic, soak up bandwidth on your uplinks and you will be left with Spanning Tree operation from core to access, giving blocked ports and potentially undeterministic behaviour.

EIGRP to the Access would allow you to load balance, making full use of all the uplinks from Access to Core, whilst also being a touch more deterministic in my book.

Just my two cents!

Regards, Ash.

[krahmani323 · Accepted solution]

Hello,
Thanks Ashley for the input.

As Ashley stated I am thinking about the broadcast/Multicast  diffusion domain induced by the use of L2 MEC (with use of trunks and  vlan interfaces). 
I would adopt L3 MEC for a domain control of the EIGRP packet exchange  (hellos, DUAL, Update....) and neighbor relationship.

Otherwise some other informations leading us to confirm L3 MEC would be the choice in our scenario :

http://www.cisco.com/en/US/products/ps9336/products_tech_note09186a0080a7c837.shtml#mec (see "MultiChassis EtherChannel" part)

* L2 MEC enables loop free topology, doubles the uplink bandwidth  as no links are blocked and provides faster convergence than STP.

* L3 MEC provides reduced neighbor counts, better load-sharing (L2 and L3 for unicast and multicast), reduced VSL link utilization for multicast flows and faster convergence than ECMP.

Hope that helps.
Regards.
Karim