What is the most appropriate network design for this topology?

[Ilya_Semenov]

Hello, everybody!

I have a task to totally redesign a huge network which consists of about 150 switches (old enterprise CISCOs). Everything is located at factory's territory. Take a look at the attached picture, this is a classic "star" topology. It's current topology.

The core of the network will be located in two separated DPC, there will be about 300 meters between them (Now it's two 6509 bounded by VSS in a same location).

I will replace all cores, distribution and 90% of access switches.

The main goal is to optimize bandwidth and increase redundancy and fail-tolerance to maximum level.

What is the most suitable topology for these goals?

Could Extreme offer something more than MLAG for core, stacks for distribution and LAGs everywhere?

I wonder how Extreme Fabric is suitable here, but there is no any success stories around...

Any ideas?

Many thanks in advance,

Ilya

[Stephane_Grosj1]

Hi,

Well, first of all, in case of dual-homing, you wouldn't be active/active, and you would need on top of MLAG to run ESRP. Also, no igmp snooping on the vlans going to VPLS, nor ipforwarding. So you'd need to dedicate some switches to perform routing, outside of the VPLS network. Is there a requirement for Multicast?

When would happen that project? If you don't want (for some reason) to use a proven Fabric technology such as Fabric Connect, there're other modern ways. But timing is important and you'd need to discuss that with your local Extreme team, that cannot happen at the moment on a public forum.

[Ilya_Semenov]

Hi, Stephane,

Budget is not a problem, feasibility is a problem.

Why "routing part of customer vlans will be complicated to manage"? Could you explain it?

I've configured MPLS only during CCIE preparation (VRF+BGP+MPLS)

Thank you!

[Stephane_Grosj1]

Building a full mesh VPLS network will be very expensive (you need MPLS everywhere, so x460G2 would be the minimal switch) and then the routing part of the customer vlans will be complicated to manage (your VPLS nodes cannot do that part - VPLS is L2 only), and you don't have active/active paths with MLAG.

If you would consider Extreme's Fabric Connect, based on SPBm and formerly from Avaya, you could build the Core of the network with that (VSP family of products), and attach to it your edge switches (via Fabric Attach). You should speak with your local SE about Fabric Attach and the plans for EXOS.

http://www.extremenetworks.com/extreme-fabric-connect/

The SPBm network would be like your VPLS idea, but better: L2 virtualization, L3 virtualization, IP Multicast virtualization, etc. There are thousands of such fabric in production. It's like an MPLS multiservices solution, but running on a single protocol, and multicast friendly.

[Ilya_Semenov]

Bin, SPB doesn't work now. No one in Extreme Networks knows how it works. You've just bought it. And there are no customers who could buy Avaya.

[Bin]

Hi, llya,

Like Daniel Flouret, you could buy Avaya switches(Extreme) now.

You could find ERS and VSP series switch from our web site.
http://www.extremenetworks.com/products/switching-routing/

Also agree with Grosjean, full mesh VPLS network will be expensive and is complex. SPBm is very simple and security. Plus, SPBm supports multiple ways for the link aggregation, such as SMLT (like MLAG).

Regards,
Bin