I looking closely at Data Centre Designs at the moment, and been considering whether a looped triangle or a looped square is the best for edge switches. I find it difficult to detect a difference that makes a clear purchasing decision. So I sat down to think about this.
The Looped Triangle provides:
- support for single switch
- simpler configuration for troubleshooting as the edge device is effectively alone. This would provide support for dual connected servers on a single switch or for non critical devices that might need only a single port.
- - less traffic in the core, servers will send keep alive packets to each other to monitor status. Having directly connected switches means that traffic will need to to the core and thus VLANs do not need to be propagated across the core switches.
- more difficult to troubleshoot, as you need to consider the other switch in the pair, especially when trying to
capture packets ro do bandwidth analysis. - less ports needed in the core

For some Data Centre designs the core will be MPLS, and I do not want VLANs to be propagated unless I am using MPLS. By allowing VLANs into the core I am effectively cerating a situation where a lazy Network Engineer can just extend VLANs and thus bypass the MPLS. In this case, the Looped Triangle is not recommended.
Consider the following MPLS core and edge for a small data centre and the VLAN traffic between what should be two PE routers.

I can still implement L2 by using L2 MPLS (in one of its many forms), but it seems to better to deliberately block VLANs to improve security and service separation. by making L3 separation the default for servers on either side of the MPLS cloud, I should be improving the future design.
For some networks, the ability to have VLANs anywhere across the network might be important, but this would remove some of the security and separation benefits of MPLS.




