Couple companies have recommended that we should build spine/leaf architecture with 1U switches. Spines with 32x 40G or 100G, leaves with 48x 10G or 25G. However we currently don't need many server facing ports as we're mainly using HCI stuff in the DC. Something like 4x48 per DC would be fine for now, though I'm not sure how much we're going to expand.
Wouldn't it be better to just get something like 2x 2U modular switch with enough ports? We're considering a 2U switch that has 4 slots, and 16x40 or 24x10/25 cards for that. As for uplinks, our campus core is with 10Gbps ports.
If we just plug in 4*24x10Gbps cards, we get 96Gbps ports. So with two switches we could attach 96 servers minus the firewalls and uplinks etc. We could expand it by getting 2 more and do full mesh. Then we could connect 192 hosts with 2x10 LACP.
Due to mergers etc. we have a lot's of smaller data centers...
Any ideas? Thanks!
Edit: forgot to add that we'd need a way to migrate stuff from DC to DC, I guess for now it's going to be VXLAN until we figure out how the applications work. And get everything redundant on the app level
No comments:
Post a Comment