Automated Campus solution also for Data Center?

  • 26 June 2019
  • 4 replies

Userlevel 3

is Automated Campus solution a possible implementation also for the Data Center in case I wanto to extend the Fabric Connect\ Fabric attach to this environment or are more the cons respect the pro in comparison to the Agile Data Center solution?
What I have less if I use the Fabric Connect solution in the datacenter (I assume that all the workload functionality will be ported to the XMC in the next months) respect the Agile Data Center?


4 replies

Userlevel 2
Hi Antonio, yes many customers are using Fabric Connect in their data centers.
What you lose compared to a BGP based fabric from Extreme is that you won't have the full DevOPs capabilities through full RestInterface support.
Also what you need to know is that east-west flow distribution is optimized for two spine nodes with FC, where BGP based fabrics are optimized for 4-8 or more spine nodes.
The rest is pretty comparable, including workload functionality and automatic VM attachments.

I hope this helps.

Userlevel 3
Thanks Roger for your reply.
Don't you think that a Fabric Connect solution can have more advantages expecially for all the best that has got this technology (single protocol, indipendent services L2, L3, DVR, best multicast scale,...) that can suggest its use also in the DC and in case of scale out with a proper topology (leaf-spine-superspine) as some topology that I've seen for example in the Extreme's campus validated design document, or in some other slides? (Instead of campus1, campus2 and campus3 in the image below, I can suppose that these are parts of the same DC or remote DC).

If I don't wrong with FC I can have up to 8 ECMP paths on higher VSP models for traffic distribution.

With Agile datacenter anyway I've got an overlay and underlay of protocols that is more similar to the solutions of the competition.
What do you think?

Userlevel 3
Currently with Extreme's Fabric Connect, we support 2 BVLANs, so between 2 given BEB nodes, traffic can take at most 2 equal cost paths.
Yes, with IP ECMP VSP can have the same IP route x8 times. So if you were learning that same IP route from say 4 different distant BEBs, then you would be able to have a x2 (BVLAN) x4 (BEBs) = x8 ECMP spread.
Increasing the number of BVLANs supported is possible and we might do so at some point; however we hardly ever have see a need for designs with more than 2 spines..
Userlevel 3
Thanks Ludovico for the explain.