Work out mpls bandwidth plan instantly with clear inputs, formula shown and shareable results.
In a hub-and-spoke MPLS design the hub circuit must carry the share of branch traffic destined for central services, discounted by how many branches are simultaneously busy. Sizing the hub at the sum of branch bandwidths is enormously wasteful; oversubscription of 3:1 to 5:1 is normal. Traffic that stays branch-to-branch is what any-to-any MPLS handles without hub capacity.
Hub sizing
aggregate branch = sites x per-site bandwidth; hub required = aggregate x hub share x simultaneity; oversubscription = aggregate / hub
Because branches are never all at peak simultaneously. Statistical multiplexing is the entire economic basis of shared WAN services.
It reverses the flow. Traffic goes to internet-hosted services rather than the data centre, which is why local internet breakout at branches has largely displaced hub backhaul.