RSTP vs Meshing in a HA Data Center

[Simon Hamilton-Wilkes]

Please share you opinions on which way to go, total domain is only 8 switches. I'm particularly interested in the effects of planned maintenance - e.g. switch software upgrades on the mesh. Servers will all be NIC teamed to two switches, other network devices will use VRRP.
TIA

Simon

[Les Ligetfalvy]

Hopefully someone will correct me if I'm wrong but I don't think you can team servers across two meshed switches.

Mesh has the advantage of link aggregation. STP kills the redundant links.

Mesh requires that all switches be at the same version. STP doesn't care.

Mesh is difficult to troubleshoot/sniff because it is proprietary and dynamic.

You might want to consider a hybrid of mesh and STP/RRP.

You could take advantage of HP network design as a complementary service.

[Kell van Daal]

If having links blocked because of spanning tree isn't a problem, then I would go with spanning tree. Either RSTP, or even MST (helps with those blocked links).
What Les says is true, this is easier to troubleshoot, and you don't have to worry about firmware versions (as much).

And I want to correct Les on the team across switches. Teaming isn't nessecarily trunking/bonding/etherchannel. There are a lot of teaming options that work independent of the switches.

[Les Ligetfalvy]

Correcting or clarifying? I am fully aware that one can team a server across switches but hold the opinion (true or false) that *meshed* switches and that sort of teaming don't go together.

[Simon Hamilton-Wilkes]

I would expect that to depend on whether you're doing failover teaming or load balancing teaming. In this small a domain how fast would you expect RSTP to converge? The application is very latency sensitive, which is why meshing is attractive to me - I'd like to eliminate even a couple of second traffic interruption. Alternative is to just plug all the systems into a single highly redundant chassis switch.

[Kell van Daal]

Guess you have to decide what is more important.
If you really want the fastest failover for the switches, meshing is the way to go.
But you will need a somewhat bigger service window for maintenance (adding a meshport requires a reboot for example).

If the service window is a problem, you may have to resort to spanning-tree.

About the teaming. Indeed what Simon says. Depends on the kind of teaming. I attached a pdf describing one team option that does work with meshes. (the whole teaming whitepaper was too big too attach)